diff --git a/.gitattributes b/.gitattributes index a974d89f33389ad8c2edd31434606574126c118c..d081c73a8e004d2c697fd6ba571ea170c02abad0 100644 --- a/.gitattributes +++ b/.gitattributes @@ -2471,3 +2471,67 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text 2025/VLOGGER_[[:space:]]Multimodal[[:space:]]Diffusion[[:space:]]for[[:space:]]Embodied[[:space:]]Avatar[[:space:]]Synthesis/e32a2703-b32a-4a6f-b45c-33f8b72e19be_origin.pdf filter=lfs diff=lfs merge=lfs -text 2025/VLog_[[:space:]]Video-Language[[:space:]]Models[[:space:]]by[[:space:]]Generative[[:space:]]Retrieval[[:space:]]of[[:space:]]Narration[[:space:]]Vocabulary/5cce3931-2141-492f-94ca-887e72e0b54b_origin.pdf filter=lfs diff=lfs merge=lfs -text 2025/VLsI_[[:space:]]Verbalized[[:space:]]Layers-to-Interactions[[:space:]]from[[:space:]]Large[[:space:]]to[[:space:]]Small[[:space:]]Vision[[:space:]]Language[[:space:]]Models/d257a4d1-091b-48d7-b804-775fddd94e7e_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VODiff_[[:space:]]Controlling[[:space:]]Object[[:space:]]Visibility[[:space:]]Order[[:space:]]in[[:space:]]Text-to-Image[[:space:]]Generation/3dac837e-c8db-4c3a-a872-8c96ea7194f5_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VSNet_[[:space:]]Focusing[[:space:]]on[[:space:]]the[[:space:]]Linguistic[[:space:]]Characteristics[[:space:]]of[[:space:]]Sign[[:space:]]Language/1135c1a4-3af0-45ac-a497-2c9557104d8c_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VTON[[:space:]]360_[[:space:]]High-Fidelity[[:space:]]Virtual[[:space:]]Try-On[[:space:]]from[[:space:]]Any[[:space:]]Viewing[[:space:]]Direction/4af7cfba-2431-4854-85f1-f3f911dcc811_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VTON-HandFit_[[:space:]]Virtual[[:space:]]Try-on[[:space:]]for[[:space:]]Arbitrary[[:space:]]Hand[[:space:]]Pose[[:space:]]Guided[[:space:]]by[[:space:]]Hand[[:space:]]Priors[[:space:]]Embedding/00462254-bb4b-4c24-9d0a-9a8f2054757a_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/V^2Dial_[[:space:]]Unification[[:space:]]of[[:space:]]Video[[:space:]]and[[:space:]]Visual[[:space:]]Dialog[[:space:]]via[[:space:]]Multimodal[[:space:]]Experts/a2fbca87-c18a-41dc-864b-f66927e9f633_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Variance-Based[[:space:]]Membership[[:space:]]Inference[[:space:]]Attacks[[:space:]]Against[[:space:]]Large-Scale[[:space:]]Image[[:space:]]Captioning[[:space:]]Models/26bae7f8-8e37-4862-840e-6578a63d1d11_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VasTSD_[[:space:]]Learning[[:space:]]3D[[:space:]]Vascular[[:space:]]Tree-state[[:space:]]Space[[:space:]]Diffusion[[:space:]]Model[[:space:]]for[[:space:]]Angiography[[:space:]]Synthesis/221d762b-beb5-4562-9aa6-3ebe0030c356_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VerbDiff_[[:space:]]Text-Only[[:space:]]Diffusion[[:space:]]Models[[:space:]]with[[:space:]]Enhanced[[:space:]]Interaction[[:space:]]Awareness/ad57dfe8-0121-4da2-a044-841b75f6e661_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/ViCaS_[[:space:]]A[[:space:]]Dataset[[:space:]]for[[:space:]]Combining[[:space:]]Holistic[[:space:]]and[[:space:]]Pixel-level[[:space:]]Video[[:space:]]Understanding[[:space:]]using[[:space:]]Captions[[:space:]]with[[:space:]]Grounded[[:space:]]Segmentation/ae891128-2b76-44ec-8a3e-060beba26fb9_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/ViKIENet_[[:space:]]Towards[[:space:]]Efficient[[:space:]]3D[[:space:]]Object[[:space:]]Detection[[:space:]]with[[:space:]]Virtual[[:space:]]Key[[:space:]]Instance[[:space:]]Enhanced[[:space:]]Network/c314dea2-42fe-4d4d-8bb4-7010c813e409_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/ViUniT_[[:space:]]Visual[[:space:]]Unit[[:space:]]Tests[[:space:]]for[[:space:]]More[[:space:]]Robust[[:space:]]Visual[[:space:]]Programming/8d692184-2a4f-4cca-806a-220832f3b327_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Vid2Avatar-Pro_[[:space:]]Authentic[[:space:]]Avatar[[:space:]]from[[:space:]]Videos[[:space:]]in[[:space:]]the[[:space:]]Wild[[:space:]]via[[:space:]]Universal[[:space:]]Prior/61a0578e-e851-4106-a096-325f4e7911db_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Vid2Sim_[[:space:]]Generalizable,[[:space:]]Video-based[[:space:]]Reconstruction[[:space:]]of[[:space:]]Appearance,[[:space:]]Geometry[[:space:]]and[[:space:]]Physics[[:space:]]for[[:space:]]Mesh-free[[:space:]]Simulation/48cd44e1-bff7-400d-aceb-8a86fa95ccb5_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Vid2Sim_[[:space:]]Realistic[[:space:]]and[[:space:]]Interactive[[:space:]]Simulation[[:space:]]from[[:space:]]Video[[:space:]]for[[:space:]]Urban[[:space:]]Navigation/a9b54426-b8df-4fc1-983c-717562672544_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VidBot_[[:space:]]Learning[[:space:]]Generalizable[[:space:]]3D[[:space:]]Actions[[:space:]]from[[:space:]]In-the-Wild[[:space:]]2D[[:space:]]Human[[:space:]]Videos[[:space:]]for[[:space:]]Zero-Shot[[:space:]]Robotic[[:space:]]Manipulation/c9488d81-95eb-4b83-bf3b-3ddc8f3b7707_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VidComposition_[[:space:]]Can[[:space:]]MLLMs[[:space:]]Analyze[[:space:]]Compositions[[:space:]]in[[:space:]]Compiled[[:space:]]Videos_/6c9efc65-d2ba-48e3-a8d2-fe7c5c1d7646_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VidHalluc_[[:space:]]Evaluating[[:space:]]Temporal[[:space:]]Hallucinations[[:space:]]in[[:space:]]Multimodal[[:space:]]Large[[:space:]]Language[[:space:]]Models[[:space:]]for[[:space:]]Video[[:space:]]Understanding/c6c9e644-f51f-4e6b-aa79-2d374850db15_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VidMuse_[[:space:]]A[[:space:]]Simple[[:space:]]Video-to-Music[[:space:]]Generation[[:space:]]Framework[[:space:]]with[[:space:]]Long-Short-Term[[:space:]]Modeling/e0ce05e4-9f59-4588-812e-59896766b6d3_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VidSeg_[[:space:]]Training-free[[:space:]]Video[[:space:]]Semantic[[:space:]]Segmentation[[:space:]]based[[:space:]]on[[:space:]]Diffusion[[:space:]]Models/a8a75be1-b6e5-46b8-a340-187193161579_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VidTwin_[[:space:]]Video[[:space:]]VAE[[:space:]]with[[:space:]]Decoupled[[:space:]]Structure[[:space:]]and[[:space:]]Dynamics/396b7efb-b08c-4897-8d56-4437ed850d42_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video[[:space:]]Depth[[:space:]]Anything_[[:space:]]Consistent[[:space:]]Depth[[:space:]]Estimation[[:space:]]for[[:space:]]Super-Long[[:space:]]Videos/3d97efd2-8c6d-451d-8d75-d43ab643dfda_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video[[:space:]]Depth[[:space:]]without[[:space:]]Video[[:space:]]Models/adc6ac4e-2f14-4dd6-b43a-b385177519a9_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video[[:space:]]Language[[:space:]]Model[[:space:]]Pretraining[[:space:]]with[[:space:]]Spatio-temporal[[:space:]]Masking/592a1785-c0f4-44c0-ad31-218ccb7da263_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video[[:space:]]Motion[[:space:]]Transfer[[:space:]]with[[:space:]]Diffusion[[:space:]]Transformers/f0f4362e-9924-4b87-aeff-45869e95fb16_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video[[:space:]]Summarization[[:space:]]with[[:space:]]Large[[:space:]]Language[[:space:]]Models/49bb8211-5bbe-4fd6-a3e5-e9eae1a8e8d9_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video-3D[[:space:]]LLM_[[:space:]]Learning[[:space:]]Position-Aware[[:space:]]Video[[:space:]]Representation[[:space:]]for[[:space:]]3D[[:space:]]Scene[[:space:]]Understanding/c7916639-0a81-44c5-a96b-970d900fe738_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video-Bench_[[:space:]]Human-Aligned[[:space:]]Video[[:space:]]Generation[[:space:]]Benchmark/b48e59c4-b36d-49ef-a7ca-02fe49c0d380_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video-ColBERT_[[:space:]]Contextualized[[:space:]]Late[[:space:]]Interaction[[:space:]]for[[:space:]]Text-to-Video[[:space:]]Retrieval/6c2b31d1-930f-426e-b6df-3bce096a9320_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video-Guided[[:space:]]Foley[[:space:]]Sound[[:space:]]Generation[[:space:]]with[[:space:]]Multimodal[[:space:]]Controls/27ba25cd-566f-43d3-badb-af29c2dceb60_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video-MME_[[:space:]]The[[:space:]]First-Ever[[:space:]]Comprehensive[[:space:]]Evaluation[[:space:]]Benchmark[[:space:]]of[[:space:]]Multi-modal[[:space:]]LLMs[[:space:]]in[[:space:]]Video[[:space:]]Analysis/8df19168-ac1f-43c7-9bef-05050364c876_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video-Panda_[[:space:]]Parameter-efficient[[:space:]]Alignment[[:space:]]for[[:space:]]Encoder-free[[:space:]]Video-Language[[:space:]]Models/8e708f5e-f4c2-49cb-b40d-85c403ad2f23_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Video-XL_[[:space:]]Extra-Long[[:space:]]Vision[[:space:]]Language[[:space:]]Model[[:space:]]for[[:space:]]Hour-Scale[[:space:]]Video[[:space:]]Understanding/410e6e98-5f80-4e5e-9672-9d5b8fa4130a_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoAutoArena_[[:space:]]An[[:space:]]Automated[[:space:]]Arena[[:space:]]for[[:space:]]Evaluating[[:space:]]Large[[:space:]]Multimodal[[:space:]]Models[[:space:]]in[[:space:]]Video[[:space:]]Analysis[[:space:]]through[[:space:]]User[[:space:]]Simulation/5a43a242-bd5d-4961-aed4-ae68b47ff795_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoComp_[[:space:]]Advancing[[:space:]]Fine-Grained[[:space:]]Compositional[[:space:]]and[[:space:]]Temporal[[:space:]]Alignment[[:space:]]in[[:space:]]Video-Text[[:space:]]Models/85b1b586-c77a-4209-812f-b601b6870f0d_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoDPO_[[:space:]]Omni-Preference[[:space:]]Alignment[[:space:]]for[[:space:]]Video[[:space:]]Diffusion[[:space:]]Generation/2f70dda9-0787-49af-abce-676cbf00394e_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoDirector_[[:space:]]Precise[[:space:]]Video[[:space:]]Editing[[:space:]]via[[:space:]]Text-to-Video[[:space:]]Models/8f2614cf-ca8d-409c-94f9-479e72f1f29f_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoEspresso_[[:space:]]A[[:space:]]Large-Scale[[:space:]]Chain-of-Thought[[:space:]]Dataset[[:space:]]for[[:space:]]Fine-Grained[[:space:]]Video[[:space:]]Reasoning[[:space:]]via[[:space:]]Core[[:space:]]Frame[[:space:]]Selection/eff006bb-cde1-4b27-a0ef-89b5fd58945a_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoGEM_[[:space:]]Training-free[[:space:]]Action[[:space:]]Grounding[[:space:]]in[[:space:]]Videos/0a3fbcb1-73fe-47da-ac43-25ceb4ef9ba0_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoGLaMM[[:space:]]_[[:space:]]A[[:space:]]Large[[:space:]]Multimodal[[:space:]]Model[[:space:]]for[[:space:]]Pixel-Level[[:space:]]Visual[[:space:]]Grounding[[:space:]]in[[:space:]]Videos/a7f6c402-bb8e-45ec-b9bf-5b07a3082f52_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoGigaGAN_[[:space:]]Towards[[:space:]]Detail-rich[[:space:]]Video[[:space:]]Super-Resolution/52ec8084-0006-41dd-a7b0-3a3880698e67_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoGuide_[[:space:]]Improving[[:space:]]Video[[:space:]]Diffusion[[:space:]]Models[[:space:]]without[[:space:]]Training[[:space:]]Through[[:space:]]a[[:space:]]Teacher's[[:space:]]Guide/8fb0f104-e76b-45ee-932c-212b94049d7b_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoHandles_[[:space:]]Editing[[:space:]]3D[[:space:]]Object[[:space:]]Compositions[[:space:]]in[[:space:]]Videos[[:space:]]Using[[:space:]]Video[[:space:]]Generative[[:space:]]Priors/2cf37ed5-ea0a-4c05-acd5-0c34fb8789ea_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoICL_[[:space:]]Confidence-based[[:space:]]Iterative[[:space:]]In-context[[:space:]]Learning[[:space:]]for[[:space:]]Out-of-Distribution[[:space:]]Video[[:space:]]Understanding/3dc9da78-cea1-471c-b4db-ee52c6853efe_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoMage_[[:space:]]Multi-Subject[[:space:]]and[[:space:]]Motion[[:space:]]Customization[[:space:]]of[[:space:]]Text-to-Video[[:space:]]Diffusion[[:space:]]Models/80815955-c6e2-4c64-a2b6-3555fa9f4dc2_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoRefer[[:space:]]Suite_[[:space:]]Advancing[[:space:]]Spatial-Temporal[[:space:]]Object[[:space:]]Understanding[[:space:]]with[[:space:]]Video[[:space:]]LLM/1eb2e581-034c-4440-aa2e-5d6157a422fc_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoSPatS_[[:space:]]Video[[:space:]]SPatiotemporal[[:space:]]Splines[[:space:]]for[[:space:]]Disentangled[[:space:]]Occlusion,[[:space:]]Appearance[[:space:]]and[[:space:]]Motion[[:space:]]Modeling[[:space:]]and[[:space:]]Editing/47539518-9a92-424c-830b-8495e559a706_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoScene_[[:space:]]Distilling[[:space:]]Video[[:space:]]Diffusion[[:space:]]Model[[:space:]]to[[:space:]]Generate[[:space:]]3D[[:space:]]Scenes[[:space:]]in[[:space:]]One[[:space:]]Step/3d9c7184-8bc7-4684-b4b3-dee716e5015a_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoTree_[[:space:]]Adaptive[[:space:]]Tree-based[[:space:]]Video[[:space:]]Representation[[:space:]]for[[:space:]]LLM[[:space:]]Reasoning[[:space:]]on[[:space:]]Long[[:space:]]Videos/40b3e4b5-66bb-4e61-88c1-0c383821b986_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VideoWorld_[[:space:]]Exploring[[:space:]]Knowledge[[:space:]]Learning[[:space:]]from[[:space:]]Unlabeled[[:space:]]Videos/d7364fb4-c126-4443-875f-c1f7daf2bf13_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Viewpoint[[:space:]]Rosetta[[:space:]]Stone_[[:space:]]Unlocking[[:space:]]Unpaired[[:space:]]Ego-Exo[[:space:]]Videos[[:space:]]for[[:space:]]View-invariant[[:space:]]Representation[[:space:]]Learning/ae8678aa-d4cb-4d54-aa26-b58e33e1e461_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/ViiNeuS_[[:space:]]Volumetric[[:space:]]Initialization[[:space:]]for[[:space:]]Implicit[[:space:]]Neural[[:space:]]Surface[[:space:]]Reconstruction[[:space:]]of[[:space:]]Urban[[:space:]]Scenes[[:space:]]with[[:space:]]Limited[[:space:]]Image[[:space:]]Overlap/23124bd5-ecfd-490d-a48b-e03c3ec06af3_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VinTAGe_[[:space:]]Joint[[:space:]]Video[[:space:]]and[[:space:]]Text[[:space:]]Conditioning[[:space:]]for[[:space:]]Holistic[[:space:]]Audio[[:space:]]Generation/a74d1a60-3b5d-4c9c-a324-7df50d716e1c_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VinaBench_[[:space:]]Benchmark[[:space:]]for[[:space:]]Faithful[[:space:]]and[[:space:]]Consistent[[:space:]]Visual[[:space:]]Narratives/e5948aac-c2d3-4df8-a7f3-82cb99be6728_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Vision-Guided[[:space:]]Action_[[:space:]]Enhancing[[:space:]]3D[[:space:]]Human[[:space:]]Motion[[:space:]]Prediction[[:space:]]with[[:space:]]Gaze-informed[[:space:]]Affordance[[:space:]]in[[:space:]]3D[[:space:]]Scenes/9b42c369-4ef9-4c45-9192-f6b0a17d8bda_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Vision-Language[[:space:]]Embodiment[[:space:]]for[[:space:]]Monocular[[:space:]]Depth[[:space:]]Estimation/959b3b25-9f17-45a6-aab1-48ac53aa864f_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Vision-Language[[:space:]]Gradient[[:space:]]Descent-driven[[:space:]]All-in-One[[:space:]]Deep[[:space:]]Unfolding[[:space:]]Networks/bd772a11-bc94-4fb3-8aee-976fd29748e0_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Vision-Language[[:space:]]Model[[:space:]]IP[[:space:]]Protection[[:space:]]via[[:space:]]Prompt-based[[:space:]]Learning/e2d53e64-2ba0-4eec-9a41-e03d1858d841_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Vision-Language[[:space:]]Models[[:space:]]Do[[:space:]]Not[[:space:]]Understand[[:space:]]Negation/f44bf4ea-b507-4a1f-8e9b-87600bcf741b_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VisionArena_[[:space:]]230k[[:space:]]Real[[:space:]]World[[:space:]]User-VLM[[:space:]]Conversations[[:space:]]with[[:space:]]Preference[[:space:]]Labels/f703962d-2329-4c48-bd31-19be8745b28f_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VisionPAD_[[:space:]]A[[:space:]]Vision-Centric[[:space:]]Pre-training[[:space:]]Paradigm[[:space:]]for[[:space:]]Autonomous[[:space:]]Driving/4e6d5045-e8c8-4646-a828-f050db70849c_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/VisionZip_[[:space:]]Longer[[:space:]]is[[:space:]]Better[[:space:]]but[[:space:]]Not[[:space:]]Necessary[[:space:]]in[[:space:]]Vision[[:space:]]Language[[:space:]]Models/d6a76b28-eeb8-41ac-bee1-8e5731b2110b_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Visual[[:space:]]Agentic[[:space:]]AI[[:space:]]for[[:space:]]Spatial[[:space:]]Reasoning[[:space:]]with[[:space:]]a[[:space:]]Dynamic[[:space:]]API/18209ccb-89a4-4fe8-8dec-31f34f20c734_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Visual[[:space:]]Consensus[[:space:]]Prompting[[:space:]]for[[:space:]]Co-Salient[[:space:]]Object[[:space:]]Detection/3f1a6dd0-b317-40d4-8fff-a7533b787af0_origin.pdf filter=lfs diff=lfs merge=lfs -text +2025/Visual[[:space:]]Lexicon_[[:space:]]Rich[[:space:]]Image[[:space:]]Features[[:space:]]in[[:space:]]Language[[:space:]]Space/18fabbbb-6482-437e-bf79-727cc6d61f01_origin.pdf filter=lfs diff=lfs merge=lfs -text diff --git a/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/3dac837e-c8db-4c3a-a872-8c96ea7194f5_content_list.json b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/3dac837e-c8db-4c3a-a872-8c96ea7194f5_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..683950865184bdabef1ddcb3732bd372567dda38 --- /dev/null +++ b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/3dac837e-c8db-4c3a-a872-8c96ea7194f5_content_list.json @@ -0,0 +1,2331 @@ +[ + { + "type": "text", + "text": "VODiff: Controlling Object Visibility Order in Text-to-Image Generation", + "text_level": 1, + "bbox": [ + 127, + 130, + 869, + 152 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Dong Liang $^{1,2}$ Jinyuan Jia $^{1,3,\\dagger}$ Yuhao Liu $^{2,\\dagger}$ Zhanghan Ke $^{2}$ Hongbo Fu $^{4}$ Rynson W.H. Lau $^{2,\\dagger}$ $^{1}$ Tongji University $^{2}$ City University of Hong Kong $^{3}$ HKUST(GZ) $^{4}$ HKUST", + "bbox": [ + 99, + 185, + 908, + 227 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "sse_liangdong@tongji.edu.cn, jinyuanjia@hkust-gz.edu.cn, yuhaoliu7456@gmail.com, zhanghake2-c@my.cityu.edu.hk, hongbofu@ust.hk, Rynson.Lau@cityu.edu.hk", + "bbox": [ + 148, + 229, + 852, + 262 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/605eb3a26f0b7f7ec9bc5363a4f2b3c1fb8f14102a074ea03cac17478070f9e8.jpg", + "image_caption": [ + "(a) Layout+Order" + ], + "image_footnote": [], + "bbox": [ + 96, + 299, + 218, + 393 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/396ae77a63730ad6b504e4a40fbeb46cb8acba828a8beb07b5bc76d08f7ca37b.jpg", + "image_caption": [ + "(b) SD 1.5 [55]" + ], + "image_footnote": [ + "\"A picture of a table, with a mug and a cup on it, the mug is on a dish.\"" + ], + "bbox": [ + 236, + 300, + 357, + 393 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/90d8390c805022ce98b0af3bd8a069e40e530d985c5c3495f8e28953b43c651b.jpg", + "image_caption": [ + "(c) GLIGEN [30]" + ], + "image_footnote": [], + "bbox": [ + 364, + 300, + 483, + 393 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/9156d041384b67c029095cc08a48ac276a6775d1663f900496069dd8f81bacb8.jpg", + "image_caption": [ + "(d) DenseDiffusion [27]" + ], + "image_footnote": [], + "bbox": [ + 488, + 300, + 614, + 393 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/3ce347d46bf265f8084f29a7a91ae70c3642042c6d1324a0846eac8aae94d125.jpg", + "image_caption": [ + "(e) MIGC [83]" + ], + "image_footnote": [], + "bbox": [ + 617, + 300, + 743, + 393 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/2e1e30381a2348527e12af9930f5f690daeb7d743f2ff622f2309da400d2597f.jpg", + "image_caption": [ + "(f) VODiff (Ours)" + ], + "image_footnote": [], + "bbox": [ + 763, + 300, + 885, + 393 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/7d2c4545c2adaf30a0ee372b5f280c38b4cfbe7e01ad5f5c7824507005d7b049.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 96, + 417, + 218, + 513 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/d002f4801e9720ae91a0bb523b77c2a4d91a051492d21f0091802344fc7b52a5.jpg", + "image_caption": [], + "image_footnote": [ + "\"A suitcase with a banana in front of it and a dog behind it.\"" + ], + "bbox": [ + 236, + 419, + 357, + 513 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/74686a116326c0a5784c05a7b5fd42838d1ce6c0c438688e1adcf37242e2f0e1.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 364, + 419, + 483, + 512 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/2c8395177c695dd085d36e42b5856418d2e985a9761daf6162d4040bdd94d995.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 488, + 419, + 614, + 512 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/ad07d422a8159eb26d5d54266d112d32414f2bd332d94e9767163a9e3dc248b6.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 617, + 419, + 743, + 513 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/0e8847e8ce613ebf337399df08f281f5452a53b69c60edaf0d07a85bf2af47d3.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 764, + 420, + 885, + 513 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/e0cd9df4466b5e25d275006dd0334a0aaafe8427b37a82dd13563634a7e6fca4.jpg", + "image_caption": [ + "Figure 1. Existing T2I methods that rely on text prompts (b) and those that combine text and layout conditioning (c-e) often struggle to produce images with accurate occlusion relationships. In this work, we propose a new framework called VODiff, which enhances control over object arrangement by introducing their visibility orders (indicated by numbers above their bounding box in the layout map in (a)) as auxiliary constraints. VODiff enables the generation of images with correct spatial arrangements and occlusion relationships." + ], + "image_footnote": [ + "\"A picture of a living room with a table and a plant in front of a sofa, and a dog on the sofa.\"" + ], + "bbox": [ + 96, + 540, + 218, + 635 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/cd961ee444dbf23977023005ab0f2dd6e8abeba45d987059b2b9ffabc9e27a6f.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 236, + 541, + 357, + 635 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/ba94a955f8739f650d20019a5b7ee45b70244a8f863f81f790cb4325b2158ee3.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 362, + 541, + 482, + 635 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/7c21e606ccfd361c39ecff52685c2c865629e60a794729a3d1840f630f373b41.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 488, + 541, + 614, + 635 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/3d4ca8009b08819f411e09f0503f73962549d923b4cc04d31f104db0bc7869c4.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 617, + 541, + 745, + 635 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/b363900ca272bb5e12e8a111fb2876bbf094d57e1f4087ab4f293a320cf2acb2.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 763, + 541, + 885, + 636 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract", + "text_level": 1, + "bbox": [ + 246, + 733, + 326, + 750 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Recent advancements in diffusion models have significantly enhanced the performance of text-to-image models in image synthesis. To enable control over the spatial locations of the generated objects, diffusion-based methods typically utilize object layout as an auxiliary input. However, we observe that this approach treats all objects as being on the same layer and neglect their visibility order, lead", + "bbox": [ + 89, + 767, + 482, + 873 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ing to the synthesis of overlapping objects with incorrect occlusions. To address this limitation, we introduce in this paper a new training-free framework that considers object visibility order explicitly and allows users to place overlapping objects in a stack of layers. Our framework consists of two visibility-based designs. First, we propose a novel Sequential Denoising Process (SDP) to divide the whole image generation into multiple stages for different objects, each stage primarily focuses on an object. Second, we propose a novel Visibility-Order-Aware (VOA) Loss to trans", + "bbox": [ + 511, + 734, + 906, + 886 + ], + "page_idx": 0 + }, + { + "type": "header", + "text": "CVF", + "bbox": [ + 106, + 2, + 181, + 42 + ], + "page_idx": 0 + }, + { + "type": "header", + "text": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore.", + "bbox": [ + 236, + 0, + 810, + 46 + ], + "page_idx": 0 + }, + { + "type": "page_footnote", + "text": "†Joint corresponding authors.", + "bbox": [ + 114, + 886, + 272, + 900 + ], + "page_idx": 0 + }, + { + "type": "page_number", + "text": "18379", + "bbox": [ + 480, + 944, + 517, + 955 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "form the layout and occlusion constraints into an attention map optimization process to improve the accuracy of synthesizing object occlusions in complex scenes. By merging these two novel components, our framework, dubbed VODiff, enables the generation of photorealistic images that satisfy user-specified spatial constraints and object occlusion relationships. In addition, we introduce VOBench, a diverse benchmark dataset containing 200 curated samples, each with a reference image, text prompts, object visibility orders and layout maps. We conduct extensive evaluations on this dataset to demonstrate the superiority of our approach.", + "bbox": [ + 88, + 90, + 480, + 257 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "1. Introduction", + "text_level": 1, + "bbox": [ + 89, + 286, + 222, + 301 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "With the advancements of diffusion models [20, 59], text-to-image (T2I) models [54, 55, 57] have revolutionized the field of image generation. These large-scale T2I models, e.g., Stable Diffusion [55], can generate diverse, high-quality images conforming to given prompts, facilitating more efficient image editing [4] and artistic creation [44].", + "bbox": [ + 89, + 311, + 482, + 402 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Despite the success, existing T2I models often struggle to produce satisfactory results in complex scenes involving multiple objects with occlusions. On the one hand, text descriptions alone lack spatial controllability, causing text-based methods to likely generate incorrect object positions and relationships, as shown in Fig. 1(b). On the other hand, although some condition-based methods [27, 30, 83] introduce layout conditions (e.g., bbox) into pre-trained T2I models to improve positional accuracy, they do not consider the occlusion relationship of objects and may produce unsatisfactory results, as shown in Fig. 1(c)-(e).", + "bbox": [ + 89, + 402, + 482, + 568 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this work, we observe that when generating a scene with multiple objects, existing methods typically employ a one-layer-for-all strategy to synthesize all objects simultaneously on a single canvas. Such a design requires the model to associate the objects with the input text description and accurately obtain their spatial/occlusion relationships from it, if available. This is inherently challenging and often results in mixed or even incorrect object arrangements. Even when layout constraints [3, 27, 47, 68, 70] are provided, the model may still struggle with synthesizing the correct visibility order of overlapping objects, resulting in inaccurate occlusions. To address this problem, we propose a new training-free Visibility Order Diffusion (VODiff) framework to explicitly model the visibility order of objects by assigning each user-specified object to a separate layer and executing a novel multi-layer denoising process to produce the output image.", + "bbox": [ + 89, + 568, + 482, + 824 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "We first propose a novel Sequential Denoising Process (SDP) to render the input objects one by one. SDP stratifies the objects into distinct layers, from bottom to top, according to the visibility order. For example, in the second example of Fig. 1, the layers are ordered with the \"dog\" at the", + "bbox": [ + 89, + 825, + 482, + 900 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "bottom, the \"suitcase\" in the middle, and the \"banana\" at the top. Second, SDP divides the complete denoising process into several sub-stages, with each sub-stage primarily focusing on generating one object. Concurrently, during the denoising process, SDP assigns dynamically varying visual guidance to different objects/layers at different stages using a newly proposed smooth guidance mechanism. To prevent the misalignment of inter-object relations caused by deviations in visual guidance during denoising, we further propose a novel Visibility-Order-Aware (VOA) loss to transform layout and occlusion constraints into attention map optimization. When synthesizing an object, we first divide the corresponding layer into three regions, overlapping, visible, and background regions, based on its visibility w.r.t. those of the other objects, and then apply different constraints on these regions.", + "bbox": [ + 511, + 90, + 903, + 332 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "To address the lack of a benchmark in visibility-order-aware T2I generation, we also curate a new benchmark called VOBench, which comprises tailored visibility orders, layouts, referenced images, and prompts. To validate the effectiveness of our VODiff, we conduct extensive evaluations on VOBench. Results show that VODiff generates prompt-consistent objects within the user-specified layout regions and precisely delineates the occlusion relationship, as shown in Fig. 1(f).", + "bbox": [ + 511, + 334, + 903, + 469 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Our key contributions can be summarized as:", + "bbox": [ + 532, + 470, + 831, + 484 + ], + "page_idx": 1 + }, + { + "type": "list", + "sub_type": "text", + "list_items": [ + "- To the best of our knowledge, we are the first to consider visibility order when generating images with T2I models. We present a new training-free framework named VODiff, which explicitly models object visibility order to control object occlusion relationships.", + "- We propose two novel designs to support the explicit modeling of object visibility order: a Sequential Denoising Process (SDP) to generate different objects sequentially at various sub-stages of the denoising process, and a Visibility-Order-Aware (VOA) Loss to incorporate occlusion relationships into the attention optimization process.", + "- We curate a visibility-order-aware benchmark called VOBench, and conduct extensive evaluations on it. Results show that compared to existing SOTA methods, VODiff can synthesize more photorealistic images, which satisfy user-specified spatial constraints as well as object visibility order." + ], + "bbox": [ + 532, + 487, + 903, + 758 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2. Related Work", + "text_level": 1, + "bbox": [ + 511, + 775, + 653, + 791 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2.1. Text-to-Image (T2I) models", + "text_level": 1, + "bbox": [ + 511, + 801, + 759, + 816 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Substantial progress has been made on T2I models through the use of diffusion models [20, 59, 60]. These models conceptualize image synthesis as an iterative denoising process guided by textual prompts. The denoising is conditioned on textual embeddings generated by language", + "bbox": [ + 511, + 824, + 903, + 900 + ], + "page_idx": 1 + }, + { + "type": "page_number", + "text": "18380", + "bbox": [ + 480, + 944, + 517, + 955 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "encoders [52, 53] and is performed either in pixel space [2, 43, 54, 57] or latent space [12, 16, 36, 48, 55, 74], followed by cascaded super-resolution [21] or latent-to-image decoding [11] for high-resolution image generation. By training on extensive datasets of image-text pairs [58], these models learn a unified generative space in which visual and linguistic modalities are closely associated.", + "bbox": [ + 89, + 90, + 480, + 196 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Nonetheless, current T2I diffusion models still struggle to accurately interpret spatial layout instructions, making it challenging to control the occlusion relationships among multiple objects using textual descriptions alone.", + "bbox": [ + 89, + 196, + 480, + 258 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.2. Layout-guided Text-to-Image (T2I) Models", + "text_level": 1, + "bbox": [ + 89, + 268, + 457, + 286 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Layout-guided T2I models aim to generate images that contain objects as specified by the input text description, with their locations aligned with the input layout.", + "bbox": [ + 89, + 292, + 480, + 337 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Training-based Models can be categorized into those that fine-tune pre-trained models [1, 7, 10, 23, 31, 51, 72, 76] using the specific input conditions, and those that train additional plug-ins [25, 30, 35, 37, 38, 40, 50, 67, 69, 79] to support various conditions. Some works achieve instance-level control [64, 65, 82, 83] by assigning individual prompts to each object to precisely control their attributes. However, these methods require additional training on large-scale datasets, resulting in significant computational and labor costs.", + "bbox": [ + 89, + 338, + 482, + 488 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Several methods [22, 26, 71, 81] employ multiple control conditions within a single generation. Sun et al. [63] address the complex relationships among different control conditions by training a multi-control encoder to maintain harmonious results when combining multiple controls. Although these approaches can generate images with harmonious occlusion relationships, they still cannot directly control the occlusion relationship of individual objects.", + "bbox": [ + 89, + 489, + 482, + 611 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Sampling-based Models often integrate multiple denoising processes, with each process targeting a specific component or region of the image. Liu et al. [33] interpret diffusion models as energy-based models and propose two compositional operators to combine the score functions of different conditions in a single sampling step. MultiDiff. [3] and LRDiff. [49] then propose to utilize shared parameters or constraints to ensure consistency/accuracy of each object in the generated image.", + "bbox": [ + 89, + 612, + 482, + 747 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Although several methods [39, 41] have proposed training-free multimodal control for the T2I generation, they primarily focus on enhancing control techniques without specifically addressing the object occlusion issues.", + "bbox": [ + 89, + 748, + 482, + 809 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Attention-based Models can be categorized into forward and backward approaches based on how they employ the attention maps. Forward methods [2, 13, 15, 27, 73] manipulate attention values directly on the feature map. The accuracy of these methods heavily depends on the precision of the layout condition (e.g., segmentation masks). Back-", + "bbox": [ + 89, + 810, + 482, + 900 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "ward methods [5, 8, 47, 68, 70, 73] combine the layout condition with the attention map, and utilize an energy function to translate the spatial constraints into the noisy input for iterative updates. These methods also primarily emphasize on positional accuracy while neglecting the occlusion relationships among objects, leading to the unsatisfactory generation of overlapping regions.", + "bbox": [ + 511, + 90, + 903, + 196 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In addition, while several customized T2I models [9, 29, 32, 61, 62, 72, 80] attempt to integrate specific input objects into designated regions of the generated image by using additional layout input, they overlook the visibility order among the integrated objects and other objects within or near those regions. This limitation hinders their effectiveness in addressing occlusion issues.", + "bbox": [ + 511, + 196, + 903, + 301 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3. Method", + "text_level": 1, + "bbox": [ + 513, + 315, + 604, + 330 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In this work, our goal is to enhance the ability of pre-trained text-to-image (T2I) diffusion models to correctly handle occlusion relationships in multi-object scenarios without training. We note that the overlapping regions of the objects are the main challenge, and they affect the occlusion relationship of the objects in the synthesized image. However, it is difficult for T2I models to understand occlusion relations among objects through text prompts only. In addition, current spatial layouts can only help constrain the spatial location of the objects. They do not indicate the order of objects in the overlapping regions.", + "bbox": [ + 511, + 340, + 903, + 507 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "To achieve our goal, we propose a training-free method, VODiff, which introduces object visibility order into the layout-guided T2I generation to effectively address the incorrect object arrangement problem. Fig. 2 shows our framework. It consists of two novel visibility-aware designs: a Sequential Denoising Process (SDP) and a Visibility-Order-Aware (VOA) loss. When generating an image, our SDP strategy divides the whole denoising process into several sub-stages according to the number of user-specified objects and generates the specified objects in each stage sequentially in a bottom-up manner. Meanwhile, at each denoising step, our VOA loss applies dynamic losses to regions with different visibility states, ensuring accurate occlusion relationships among the generated objects.", + "bbox": [ + 511, + 507, + 903, + 718 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "For the rest of this section, we first briefly introduce the preliminary information of diffusion models and attention layers in Sec. 3.1. We then present how we design the SDP in Sec. 3.2 and detail the VOA loss function in Sec. 3.3.", + "bbox": [ + 511, + 718, + 903, + 779 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.1. Preliminaries", + "text_level": 1, + "bbox": [ + 511, + 787, + 653, + 801 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Diffusion Models [20, 59, 60] learns to approximate a data distribution by progressively denoising a random variable sampled from a unit Gaussian distribution $\\mathcal{N}(0,I)$ . At a denoising step $t\\in [1,\\dots ,T]$ , the forward-time Markov chain is formulated as $x_{t} = \\sqrt{1 - \\beta_{t}} x_{t - 1} + \\sqrt{\\beta_{t}}\\epsilon_{t - 1}$ , where $\\epsilon \\sim \\mathcal{N}(0,I)$ , and $\\beta_{t}$ is a pre-defined variance schedule.", + "bbox": [ + 511, + 809, + 903, + 900 + ], + "page_idx": 2 + }, + { + "type": "page_number", + "text": "18381", + "bbox": [ + 480, + 944, + 517, + 955 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/2e4359b52be4386cfa615893b5865b4fff7bc1f690515d4dacbb3bb7a3af8615.jpg", + "image_caption": [ + "Figure 2. Our VODiff incorporates two visibility-aware designs: a sequential denoising process (SDP) and a visibility-order-aware (VOA) loss function. SDP divides the denoising process into several stages according to the visibility order, and performs denoising on different parts sequentially. Meanwhile, the VOA loss is utilized to eliminate the local spatial shifts in the attention layers, further preventing position deviations and incorrect occlusion relationships. Given a text prompt, the spatial layout, and the visibility order of the objects to be synthesized as inputs, our VODiff can generate images that simultaneously adhere to all the constraints." + ], + "image_footnote": [], + "bbox": [ + 117, + 84, + 883, + 311 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The reverse-time denoising process can then be defined as: $x_{t - 1} = \\tilde{\\alpha}_t x_t + \\tilde{\\beta}_t \\hat{s}_t + \\sigma_t \\epsilon_t$ , where $\\tilde{\\alpha}_t, \\tilde{\\beta}_t$ , and $\\sigma_t$ are calculated coefficients. To generate an image, given a noisy input $x_T$ sampled from $\\mathcal{N}(0, I)$ , we run denoiser $\\epsilon_\\theta$ iteratively to obtain intermediate representations $x_{T - 1}, \\dots, x_1, x_0$ where $x_0 \\approx x$ is the final result.", + "bbox": [ + 89, + 400, + 483, + 491 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Attention Layers are the cornerstone of Stable Diffusion [55]. They used to condition the generation based on the text prompt. A pretrained CLIP encoder [52] is usually utilized to encode the text prompt $c = (c_{1}, c_{2}, \\dots, c_{n})$ , where $n$ is the number of tokens, into text embeddings. These tokens are then used as keys $K \\in \\mathbb{R}^{n \\times d}$ and values $V \\in \\mathbb{R}^{n \\times d}$ in the Cross-Attention layer. The symbol $d$ denotes the feature dimension. For a set of queries $Q \\in \\mathbb{R}^{hw \\times d}$ computed from the feature map of size $h \\times w$ , the cross-attention map $A_{t}$ at time step $t$ is calculated as: $A_{t} = \\text{softmax}\\left(\\frac{QK^{\\top}}{\\sqrt{d}}\\right) \\in [0, 1]^{hw \\times n}$ , where $A_{t}$ contains $n$ attention maps $\\{A_{t}^{1}, \\dots, A_{t}^{n}\\}$ , and $A_{t}^{i} \\in [0, 1]^{hw}$ indicates the strength of association between each word token $c_{i}$ and each spatial location on the feature map. A higher attention value in $A_{t}^{i}$ indicates a higher probability of generating the object $c_{i}$ at that spatial location.", + "bbox": [ + 89, + 492, + 483, + 742 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.2. Sequential Denoising Process (SDP)", + "text_level": 1, + "bbox": [ + 89, + 756, + 401, + 772 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Instead of directly generating all objects on a single layer at once, like what existing works [27, 39, 47] do, we are inspired by the painter's algorithm [42], which sorts all polygons according to their z values and then render them starting from the farthest polygon, and the layer concept in image editing tools such as Photoshop. We propose to divide the generation of the target scene into layers, where each object is assigned a separate layer, and synthesize the ob", + "bbox": [ + 89, + 780, + 483, + 902 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "jects sequentially, layer by layer from bottom to top, with the following steps.", + "bbox": [ + 511, + 400, + 903, + 430 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "First, we stratify the target objects into distinct layers according to their visibility order. Specifically, given an input text prompt $C$ containing $N$ target objects; a layout condition represented by bounding boxes $B = \\{b_{1}, b_{2}, \\ldots, b_{N}\\}$ , where $b_{n}$ is the bounding box of the $n$ -th target object; and visibility orders $V = \\{v_{1}, v_{2}, \\ldots, v_{N}\\}$ , where $v_{n}$ is the visibility order of the $n$ -th target object, we convert the layout condition $B$ into $N$ masks, $M = \\{m_{1}, m_{2}, \\ldots, m_{N}\\}$ , where $m_{n}$ is the mask corresponding to the bounding box $b_{n}$ with a value of 1 inside the bounding box region and 0 elsewhere. Meanwhile, we extract $N$ object entities from the input text prompt $C$ and denote the set as $\\{c_{1}, c_{2}, \\dots, c_{N}\\}$ . Each bounding box $b_{n}$ is associated with an object prompt $c_{n}$ , which describes the content of the target object inside $b_{n}$ . We order the $N$ target objects in an increasing object visibility order specified by the user, even though there may be non-overlapping objects. The layers from bottom to top can thus be represented as $L = \\{(m_{1}, c_{1}, v_{1}), (m_{2}, c_{2}, v_{2}), \\dots, (m_{N}, c_{N}, v_{N})\\}$ .", + "bbox": [ + 511, + 431, + 906, + 719 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Second, we divide the reverse-time denoising process (with a total of $T$ steps) of our SDP into two main parts, $[T, T_g]$ and $[T_g, 0]$ , where $T_g$ denotes the start time of global denoising. The global denoising is to delineate the overall scene based on the input text prompt $C$ . The first part of our SDP focuses on per-layer generation based on layers $L$ . We divide it into $N$ sub-stages, $S = \\{s_1, s_2, \\dots, s_N\\}$ , starting from the bottom layer (i.e., the object with the lowest visibility order). Each sub-stage $s_n$ contains $k$ denoising steps, ranging from $[s_n, s_n + k]$ , i.e., $T - T_g = kN$ . Since the diffusion denoising process is reversed, $s_n + k$ represents the start time for substage $s_n$ . At each sub-stage, to", + "bbox": [ + 511, + 719, + 908, + 901 + ], + "page_idx": 3 + }, + { + "type": "page_number", + "text": "18382", + "bbox": [ + 480, + 944, + 517, + 955 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "force $c_{n}$ to be generated within the object region, we explicitly incorporate visual guidance into this region. However, we observe that focusing exclusively on a single object $n$ at each stage can lead to the failure of generating other objects, particularly those whose start time is near $T_{g}$ , due to the missing of structure information [2] at later denoising steps. To address this problem, at each stage $n$ , we apply visual guidance not only for the target object $n$ but also for other objects $j$ ( $j \\neq n$ ), to ensure the generated layout adheres to the overall requirements. Our scheme for adding visual guidance can be formulated as:", + "bbox": [ + 88, + 90, + 482, + 257 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nx _ {t, n} = x _ {t} + \\lambda_ {t, n} \\odot m _ {n}, \\tag {1}\n$$\n", + "text_format": "latex", + "bbox": [ + 202, + 263, + 482, + 279 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $\\lambda_{t,n}$ represents the visual guidance scale of the target object $n$ at time step $t$ . Note that we allocate a dynamically varying smooth $\\lambda_{t,n}$ to each object to ensure correct occlusion relationships. Specifically, when processing object $n$ at sub-stage $n$ , the SDP assigns a guidance scale $a$ to object $n$ and $a/2$ to all other objects. As the process transitions from sub-stage $n$ to sub-stage $n + 1$ , the guidance scale for object $n$ decreases linearly to $a/2$ , while the guidance scale for object $n + 1$ increases linearly to $a$ , and guidance scale for the other objects remains unchanged, as:", + "bbox": [ + 89, + 282, + 483, + 434 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\n\\lambda_ {t, n} = \\left\\{ \\begin{array}{l l} a - \\frac {a}{2} \\frac {t - s _ {n - 1}}{k} & \\text {i f} s _ {n - 1} < t < = s _ {n - 1} + k, \\\\ \\frac {a}{2} + \\frac {a}{2} \\frac {t - s _ {n}}{k} & \\text {i f} s _ {n} < t < = s _ {n} + k, \\\\ \\frac {a}{2} & \\text {o t h e r w i s e}, \\end{array} \\right. \\tag {2}\n$$\n", + "text_format": "latex", + "bbox": [ + 102, + 438, + 482, + 494 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $a$ denotes the initial visual guidance scale (refer to Sec. D.4 of the Supp. for more discussion) and $s_0 = T$ . By assigning a smoothly transitioning visual guidance, we ensure that each sub-stage focuses more on a specific object while maintaining correct occlusion relationships between it and other objects, such that in overlapping regions, objects with a higher visibility order occlude those with a lower one. Finally, we forward the updated noisy input $x_{t,n}$ to the denoising model $\\epsilon_{\\theta}$ to estimate noise for each object separately and then combine them together, as:", + "bbox": [ + 89, + 496, + 483, + 647 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\n\\epsilon_ {\\theta} \\left(x _ {t}, t, C\\right) = \\sum_ {n = 1} ^ {N} \\frac {m _ {n}}{\\sum_ {n = 1} ^ {N} m _ {n}} \\otimes \\epsilon_ {\\theta} \\left(x _ {t, n}, t, c _ {n}\\right), \\tag {3}\n$$\n", + "text_format": "latex", + "bbox": [ + 116, + 651, + 482, + 691 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $\\epsilon_{\\theta}(x_t,t,C)$ is the estimated noise used to update $x_{t - 1}$ via the standard diffusion scheduler.", + "bbox": [ + 89, + 696, + 482, + 726 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.3. Visibility-Order-Aware (VOA) Loss", + "text_level": 1, + "bbox": [ + 89, + 734, + 398, + 752 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Although SDP can help constrain target generation within the specified region using visual guidance, the consecutive conv. layers of the denoising model introduce local shifts, which are then amplified by the attention layers. As a result, adding visual guidance at the input level (i.e., $x_{t}$ ) as a constraint is insufficient and tends to cause object position deviation and false occlusion relationships (see Fig. 5 column 2 for a visual illustration).", + "bbox": [ + 89, + 757, + 483, + 878 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/0726e66f0937cde8976af8f2f45d781c5d3838e665730c3ac944c38a82b01c79.jpg", + "image_caption": [ + "Layout + Order" + ], + "image_footnote": [], + "bbox": [ + 521, + 90, + 638, + 170 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/b7b3e19830a58b789a6962629bf870cdb82bf3b59d89d78141f629252401ffba.jpg", + "image_caption": [ + "moverl mvisible mbackgrd", + "Figure 3. Region separation illustration in the VOA loss. $m_{n}^{\\mathrm{overlap}}$ represents the regions where the current object $n$ occludes other objects, $m_{n}^{\\mathrm{visible}}$ and $m_{n}^{\\mathrm{background}}$ denote where the current object should and should not be generated in the final image." + ], + "image_footnote": [], + "bbox": [ + 651, + 90, + 769, + 172 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/76a7c050cf5dd8b17b460fa1a69427b6309273834ae26436aa2950603ffac531.jpg", + "image_caption": [ + "Output Image" + ], + "image_footnote": [], + "bbox": [ + 781, + 90, + 897, + 171 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "To address the above problem, we propose constraining the spatial location and occlusion relationship at feature-level. Specifically, when generating an image with multiple target objects, different object prompts usually have different activations in different regions of the cross-attention map. At a given spatial location, the higher the attention value of a prompt, the more likely the corresponding object will be generated in that region [17]. In addition, when two objects overlap each other, their occlusion relationship is directly influenced by their visibility order. Once the order changes, the content of the overlapping region changes accordingly. Based on this, we transform the spatial layout and occlusion relationship among the objects to the cross-attention map and optimize it for correct image generation.", + "bbox": [ + 511, + 260, + 905, + 470 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "As shown in Fig. 3, when processing target object $n$ , we first divide the mask $m_{n}$ into three parts, overlapping regions $m_{n}^{\\mathrm{overlap}}$ , visible regions $m_{n}^{\\mathrm{visible}}$ , and background regions $m_{n}^{\\mathrm{background}}$ , based on the visibility order and masks of all objects in the layer set $L$ . Specifically, for $m_{n}^{\\mathrm{overlap}}$ , we compute the overlapping regions (i.e., dark orange region) of the current object mask $m_{n}$ and the masks of all objects with lower visibility order than the current object (i.e., $\\{m_{j}\\}$ , with $j \\in [1, n-1]$ ) as $m_{n}^{\\mathrm{overlap}} = \\bigcup_{j=1}^{n-1} m_{n} \\odot m_{j}$ . This is because an object located at an upper layer should obtain a higher attention score in the overlapping regions. For $m_{n}^{\\mathrm{visible}}$ , we first compute the overlapping regions of the current mask $m_{n}$ with the masks of all objects that have a higher visibility order (i.e., $\\{m_{j}\\}$ , with $j \\in [n+1, N]$ ). We then subtract these overlapping regions from the current object mask, which can be expressed as $m_{n}^{\\mathrm{visible}} = m_{n} - \\bigcup_{j=n+1}^{N} m_{n} \\odot m_{j}$ . $m_{n}^{\\mathrm{background}}$ is then obtained by inverting $m_{n}^{\\mathrm{visible}}$ as $m_{n}^{\\mathrm{background}} = 1 - m_{n}^{\\mathrm{visible}}$ .", + "bbox": [ + 511, + 472, + 905, + 751 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "During the denoising phase for the $n$ -th target object at time step $t$ , as indicated in Eq. 3, we extract the cross-attention maps referred to by the object prompt $c_{n}$ from each layer and average them to produce $\\mathbf{A}_{t,n}$ . We then compute the loss between the attention map $\\mathbf{A}_{t,n}$ and each of the regions. For the overlapping regions, which are affected by the visibility order, we ensure that objects in the upper layers obtain higher attention scores, as:", + "bbox": [ + 511, + 751, + 905, + 871 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\n\\mathcal {L} _ {t, n} ^ {\\mathrm {o}} = \\alpha_ {o} \\cdot \\left[ 1 - \\min \\left(\\mathbf {A} _ {t, n} \\odot m _ {n} ^ {\\text {o v e r l a p}}\\right) \\right]. \\tag {4}\n$$\n", + "text_format": "latex", + "bbox": [ + 571, + 883, + 903, + 902 + ], + "page_idx": 4 + }, + { + "type": "page_footnote", + "text": "For simplicity, we omit the classifier-free guidance [19] here.", + "bbox": [ + 112, + 886, + 437, + 900 + ], + "page_idx": 4 + }, + { + "type": "page_number", + "text": "18383", + "bbox": [ + 480, + 944, + 517, + 955 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/f286221b7c77d1d015e6b94307e065093d084e920aad7a540bd50ee82e31d13a.jpg", + "image_caption": [ + "\"A picture of a cat, in front of a CD.\"" + ], + "image_footnote": [], + "bbox": [ + 99, + 85, + 197, + 157 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/ce17ba2ab414c47029b14a0a20f03afb96287a2ec61845bd18ae535b319a9614.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 207, + 85, + 305, + 157 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/9d734215b5e77186a88ff5776f6b8b030ff6c2c7c48e1b2eb44940b5c2ef9eb3.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 307, + 85, + 400, + 157 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/ae0960715533bf88598d93aa1a4b70a2e58a5b9d23565941ed7ba18916b0665f.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 401, + 85, + 498, + 157 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/9df1fea1ee6af4ad2d64997fcbd1a400e5f7a3ca3fb6bd32d23023877c492a3a.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 501, + 85, + 599, + 157 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/da0711edb9b22e5fdf281266ad5ec27723e5eceeb5cec78dccb2430f09990a8f.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 604, + 85, + 700, + 157 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/b31f5795205c9628c116d0e5ec8c6ccd4c8ae76d331d965d80864c8ea4538fb7.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 704, + 85, + 800, + 157 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/ba6d297d1203906f1275aa759b9b5eb5e6079fedfe8cbecfe9d52c6cdab61c4e.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 803, + 85, + 900, + 157 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/74a4e7daffbb1f5605e43931c895c8613910e365c9a98c008e380c9e4a5ec4f0.jpg", + "image_caption": [ + "\"A laptop on the table with a plant behind it, and a cup near it.\"" + ], + "image_footnote": [], + "bbox": [ + 101, + 175, + 199, + 250 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/063d3786b086993b97a559f2741de26895aec842663364a2495a8ef5dbe6da10.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 207, + 174, + 305, + 247 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/563b95716ece1982d690cc18978f54f7343da30aa3588490567b348b7d60b8ed.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 307, + 174, + 400, + 247 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/0ee7477788993e3024b0a376df272a5d9cef0127487f94255cadb7092d74d8ba.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 401, + 174, + 498, + 247 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/be4407ad3e94c30cddaea4eda069f97ac82efc83503b3d7b739568a267444da5.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 501, + 174, + 598, + 247 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/2edd5e00bffc71990355a2151cd9a70192563c824b967ef62ed8d8e5878a8756.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 604, + 174, + 700, + 247 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/21dd74696ef2ff62afd698794830001a938fa8943442d8ec0f388e11b3a7fcf0.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 704, + 174, + 800, + 247 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/7e3083d800d017630758cd36e8de6ab22b08836b478e4a84ee7f9aff1ab6c428.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 803, + 174, + 900, + 247 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/b280df6c33bf337eb85bf78871cc45a18bbc7e12810b012de9d1b3af6746ca54.jpg", + "image_caption": [ + "\"Left chair and right chair, a lamp and a plant in front of a bookshelf.\"" + ], + "image_footnote": [], + "bbox": [ + 101, + 267, + 197, + 340 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/582497669611b385566a15c4e9b3e0e11549314bd54777fb6575cb48db381b10.jpg", + "image_caption": [ + "Figure 4. Qualitative comparison with six SOTA methods in Tab. 1. Training-based methods are labeled with $\\dagger$ ." + ], + "image_footnote": [], + "bbox": [ + 207, + 268, + 305, + 339 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/657650ba2ba1848ac104b557bea57d507980cc91ad17e2655f20a7914cc106f4.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 307, + 268, + 400, + 339 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/3d2e2486f491fba8ad002fc48000a214a3012963043f68fa76f9ec785fba5d19.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 401, + 268, + 498, + 339 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/afeaeed0f2717ad1f2d0133eb34d2c20d8ed2957758e4817b3180e49a6e86e4b.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 501, + 268, + 599, + 339 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/2613240aa8c6356dd8c6dc2ff6cc69d524b1762a3f90e9205cd64cddc8d06d93.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 604, + 268, + 700, + 339 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/d91a5301d1caf3d62be6c3decee53a5b22e71be91798fc9d6818783dc72f24ba.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 704, + 268, + 800, + 339 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/e1c8306413edda32a103e8f9d238359270439bbeb8079cf3253c0cb0a13dbe7e.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 803, + 268, + 900, + 339 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We then encourage target object $n$ to be generated within the visible regions by increasing the attention score through $\\mathcal{L}_{t,n}^{\\mathrm{v}}$ , and prevent the target deviation by weakening the attention score of the background regions via $\\mathcal{L}_{t,n}^{\\mathrm{b}}$ :", + "bbox": [ + 89, + 401, + 483, + 463 + ], + "page_idx": 5 + }, + { + "type": "equation", + "text": "\n$$\n\\mathcal {L} _ {t, n} ^ {\\mathrm {v}} = \\alpha_ {v i s} \\cdot \\left[ 1 - \\max \\left(\\mathbf {A} _ {t, n} \\odot m _ {n} ^ {\\text {v i s i b l e}}\\right) \\right], \\tag {5}\n$$\n", + "text_format": "latex", + "bbox": [ + 143, + 470, + 482, + 489 + ], + "page_idx": 5 + }, + { + "type": "equation", + "text": "\n$$\n\\mathcal {L} _ {t, n} ^ {\\mathrm {b}} = \\alpha_ {b a c} \\cdot \\max \\left(\\mathbf {A} _ {t, n} \\odot m _ {n} ^ {\\text {b a c k g r o u n d}}\\right), \\tag {6}\n$$\n", + "text_format": "latex", + "bbox": [ + 155, + 491, + 482, + 510 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $\\alpha_{o}$ , $\\alpha_{vis}$ and $\\alpha_{bac}$ are coefficients for the three loss functions. Finally, the total loss is:", + "bbox": [ + 89, + 513, + 482, + 542 + ], + "page_idx": 5 + }, + { + "type": "equation", + "text": "\n$$\n\\mathcal {L} _ {t} = \\sum_ {n} ^ {N} \\left(\\mathcal {L} _ {t, n} ^ {\\mathrm {o}} + \\mathcal {L} _ {t, n} ^ {\\mathrm {v}} + \\mathcal {L} _ {t, n} ^ {\\mathrm {b}}\\right). \\tag {7}\n$$\n", + "text_format": "latex", + "bbox": [ + 183, + 542, + 480, + 578 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "With the visibility-order-aware loss, we modify the noise sample $x_{t}$ at each denoising step to minimize the loss using gradient descent $\\hat{x}_{t} \\gets x_{t} - \\alpha \\nabla_{x_{t}}\\mathcal{L}_{t}$ , where $\\alpha$ controls the influence of the optimization in the denoising process. In each step $t$ , we iterate the gradient descent process $\\tau$ times. By applying the VOA loss on all objects at each sub-stage, VODiff alleviates the guidance deviation and achieves more accurate object spatial position and occlusions.", + "bbox": [ + 89, + 578, + 483, + 698 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4. Experiment", + "text_level": 1, + "bbox": [ + 89, + 703, + 214, + 719 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.1. Metrics", + "text_level": 1, + "bbox": [ + 89, + 727, + 187, + 742 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We adopt a few metrics to evaluate our model:", + "bbox": [ + 89, + 750, + 398, + 763 + ], + "page_idx": 5 + }, + { + "type": "list", + "sub_type": "text", + "list_items": [ + "- FID: We use Fréchet Inception Distance [18] to evaluate the image quality. It compares the distribution of generated images to that of real images. A lower FID score indicates a higher image quality, reflecting closer alignment with real images on visual features and diversity.", + "- CLIP-Score: We use the CLIP-Score to compute the distance between input textual features and generated image features, evaluating the fidelity of the generated image to the textual input." + ], + "bbox": [ + 89, + 763, + 483, + 900 + ], + "page_idx": 5 + }, + { + "type": "list", + "sub_type": "text", + "list_items": [ + "- **Layout Alignment (LA):** We use Grounding-DINO [34] to assess Layout alignment accuracy. It detects bounding boxes for each instance and then compute the mean IoU between the detected and Ground Truth boxes. The box threshold is set to 0.5 to avoid incorrect box detection.", + "- Occlusion Alignment (OA): We employ GPT-4V to determine the occlusion relationships between objects in the generated images by prompting it with the image and the object names contained in the image. We then compare the visual order returned by GPT-4V with the ground truth visibility order. If the two match with each other, we count it as accurate. Finally, we compute the percentage of accurate occlusion relationships from all the samples." + ], + "bbox": [ + 513, + 401, + 905, + 599 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.2.VObench", + "text_level": 1, + "bbox": [ + 511, + 607, + 624, + 622 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Since there are no visibility-order-aware datasets available for evaluation, we curate a VOBench for this purpose. All the raw image data in VOBench are manually collected from the Internet. The data annotation process is as follows. We first employ GroundingDINO [34] and GPT-4V to generate the bounding boxes and text prompts for each object in the image, and manually assign the visibility order to each object box. Based on the number of boxes detected by GroundingDINO, we then divide the images into four categories containing 2-5 objects. For each category, we manually select 50 images based on the degree of overlap between boxes to make sure that various occlusion relations are in our VOBench.", + "bbox": [ + 511, + 626, + 906, + 821 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.3. Comparison with State-of-the-Art Methods", + "text_level": 1, + "bbox": [ + 511, + 832, + 880, + 848 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We compare VODiff with 12 layout-guided text-to-image generation methods, including seven training-free methods: BoxDiff [70], R&B [68], Attention-Refocusing [47],", + "bbox": [ + 511, + 854, + 905, + 900 + ], + "page_idx": 5 + }, + { + "type": "header", + "text": "(a) Layout+Order", + "bbox": [ + 107, + 66, + 194, + 80 + ], + "page_idx": 5 + }, + { + "type": "header", + "text": "(b) InstanceDiff.† [65] (c) AnyControl† [63](d) ControlNet† [79]", + "bbox": [ + 200, + 66, + 503, + 82 + ], + "page_idx": 5 + }, + { + "type": "header", + "text": "(e) MIGC† [83]", + "bbox": [ + 516, + 66, + 594, + 80 + ], + "page_idx": 5 + }, + { + "type": "header", + "text": "(f) RPG-Diff. [73]", + "bbox": [ + 609, + 68, + 696, + 80 + ], + "page_idx": 5 + }, + { + "type": "header", + "text": "(g) FreeControl [39]", + "bbox": [ + 707, + 68, + 803, + 80 + ], + "page_idx": 5 + }, + { + "type": "header", + "text": "(h) Ours", + "bbox": [ + 828, + 68, + 870, + 80 + ], + "page_idx": 5 + }, + { + "type": "page_number", + "text": "18384", + "bbox": [ + 480, + 944, + 517, + 955 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/5c3d44859523e0d83c5ee910e436a5db305ace238c3fe3c72931aa73778a3dcc.jpg", + "table_caption": [ + "Table 1. Quantitative comparison of our VODiff with training-based and training-free methods. LA and OA represent the layout and occlusion alignment accuracy, respectively. AR denotes the average ranking from the user study, while Q, L, and O indicate that the ranking is based on Quality, Layout, and Occlusion. The best and second-best performances are marked in bold and underlined, respectively." + ], + "table_footnote": [], + "table_body": "
MetricsTraining-basedTraining-free
SmartCtrl [35]ControlNet [79]MIGC[83]InstanceDiff. [65]AnyControl [63]BoxDiff [70]R&B [68]AR [47]MultiDiff. [3]DenseDiff. [27]FreeCtrl [39]RPG [73]Ours
FID ↓10.5818.919.8210.2110.3521.9912.3714.8716.3421.6817.3511.6610.03
CLIP-Score ↑29.4529.3429.6129.5329.6329.4628.7229.5927.4229.3329.1229.5729.73
LA ↑54.8152.7354.6253.5854.7713.5030.8829.7916.6024.1136.1451.1555.11
OA ↑55.0051.5065.0063.5069.5018.5034.5032.0021.5024.0028.5051.5082.50
AR-Q ↓5.3312.132.476.203.058.1510.054.1312.687.109.005.121.50
AR-L ↓2.166.214.045.113.1212.989.0210.0712.0311.148.087.171.73
AR-O ↓6.255.123.184.022.0512.749.0110.1112.0411.098.037.101.26
", + "bbox": [ + 98, + 135, + 893, + 242 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/4f63dad8f4729f4a8ba2d36b69ab0962d834741367c7df12bf9bcedef1b286e6.jpg", + "table_caption": [ + "Table 2. Ablation study on components. Constant and Dynamic represent the state of change of guidance scale in the SDP." + ], + "table_footnote": [], + "table_body": "
SDP \nConstant DynamicVOAFID ↓Clip-Score ↑LA ↑OA ↑
✓9.9827.8532.3354.50
✓13.3429.0143.2573.50
✓✓11.6829.3343.3669.50
✓✓10.0329.7355.1182.50
", + "bbox": [ + 104, + 287, + 464, + 400 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "MultiDiffusion [3], DenseDiffusion [27], FreeControl [39] and RPG-Diffusion [73], and five training-based methods: SmartControl [35], ControlNet [79], AnyControl [63], MIGC [83] and InstanceDiffusion [65]. Since these methods do not support visibility order as an input, we include descriptions of object occlusion relationships in the text prompts to ensure fairness.", + "bbox": [ + 89, + 417, + 483, + 523 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Qualitative Comparison. As shown in Fig. 4, we visually compare VODiff with six top-performing methods. The images generated by VODiff exhibit superior spatial consistency and occlusion handling. Specifically, our method handles varying numbers of objects and degrees of occlusion remarkably well. In contrast, other methods exhibit issues such as incorrect spatial positioning (e.g., the 2nd row, (b-f)) or missing objects (e.g., the 3rd row, (d-f)). In addition, although some methods manage to generate correct spatial relationships, they fail in occlusion handling (e.g., the 1st row, (b,e,f,g)).", + "bbox": [ + 89, + 530, + 483, + 696 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Quantitative Comparison. As shown in Tab. 1, we evaluate our method against both training-free and training-based approaches. Our method significantly outperforms the training-free methods across all metrics. We achieve higher CLIP-Score (29.73), LA (55.11), and OA (82.50) scores, indicating superior image quality, layout alignment, and occlusion handling. Despite being training-free, our method outperforms training-based methods in six out of seven metrics. We achieve higher CLIP-Score (29.73 vs 29.63 by AnyControl), LA (55.11 vs 54.81 by SmartControl), and OA (82.50 vs 69.50 by AnyControl) scores. This demonstrates the effectiveness of our method in generating text-/layout-aligned and occlusion handled images without", + "bbox": [ + 89, + 704, + 483, + 900 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/cecbb91620f2c77915d98270bf5fde2d7c24e901ab3f4010a80789567c45d16f.jpg", + "image_caption": [ + "\"A bowl of strawberries and a cup on a blanket\"" + ], + "image_footnote": [], + "bbox": [ + 527, + 273, + 602, + 321 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/41c8934317f3065450f9d280f8e83029236964029378cae0582e953881982acd.jpg", + "image_caption": [ + "Layout + Order", + "SDP" + ], + "image_footnote": [], + "bbox": [ + 604, + 271, + 673, + 321 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/f8d3e1200333f75876254ebc53169bda58386ae479f89ca832f206c1de7f6068.jpg", + "image_caption": [ + "VOA" + ], + "image_footnote": [], + "bbox": [ + 676, + 271, + 746, + 321 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/3bac4bcb48d2485fd1ec6048dd06ef7ea760667f5fd8498758ef20b9962b4110.jpg", + "image_caption": [ + "VOA+SDPc" + ], + "image_footnote": [], + "bbox": [ + 750, + 271, + 818, + 321 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/2e8af02634b88617c21adff2ce9baaeb55674dacddc9f187f1cd5a5d25c3caab.jpg", + "image_caption": [ + "Ours" + ], + "image_footnote": [], + "bbox": [ + 820, + 271, + 890, + 321 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/03374d504f92570bcf9c0141629f3e33f3305ce8cf55a682808345a9935959b1.jpg", + "image_caption": [ + "\"A man and a dog in front of a sofa\"", + "Figure 5. Visual comparison of the ablation study in Tab. 2. $\\mathrm{SDP}_c$ denotes that the constant guidance scale is used. By default, the dynamic guidance scale is used in SDP." + ], + "image_footnote": [], + "bbox": [ + 527, + 335, + 599, + 386 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/299d32385b9527f212ba2dea35a6140e9074ea35062a227b4059664b609b1b09.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 604, + 335, + 673, + 386 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/997686e00c4a7784e1cdeb12ce40b00a953f11cf87dcc97cb21ebaf6cf1824d0.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 676, + 335, + 745, + 386 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/31d3aaf85111125ebeaf052529692cc6c988e95a07f05d51e6f1c22d159fafbc.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 750, + 335, + 816, + 386 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/8fa3e6cd82c0e860c844c50297115a37476eb73992c9f004202559cbe3ccce5d.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 821, + 335, + 888, + 386 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "additional training. Although the training-based method, MIGC, achieves the best FID score of 9.82, our method achieves a similarly FID of 10.03. The slight reduction can be attributed to the optimization on attention maps [47].", + "bbox": [ + 511, + 460, + 905, + 521 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "User Study. We also conduct a user study with 63 participants to assess various methods through subjective evaluation. We present users with ten sets of 8 images along the same input conditions (i.e., prompt and layout map), and ask them to rank the given images based on image quality (AR-Q), layout (AR-L), and occlusion accuracy (AR-O). Each pair of images are shown randomly, and we collect ranking scores from different participants. We then compute the average ranking (AR) for three aspects: quality, layout, and occlusion as shown in Tab. 1. The results show a clear preference for our method among the users.", + "bbox": [ + 511, + 521, + 905, + 686 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.4. Ablation Study", + "text_level": 1, + "bbox": [ + 511, + 689, + 666, + 704 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Component Analysis. We ablate our VODiff by removing SDP and VOA separately, and replacing the smooth guidance with constant guidance (denoted as $\\mathrm{VOA + SDP}_c$ ). Tab. 2 and Fig. 5 show the quantitative and qualitative results. We have three conclusions. Using only SDP (Dynamic) results in the worst LA, OA accuracy and Clip-Score but produces better FID. This is because SDP only adds visual guidance to guide the denoising direction, which does not impose stronger spatial constraints, and thus does not significantly affect the denoising process. VOA significantly improves the layout and occlusion accuracy by optimizing the cross-attention map, but at the same time, causes the model to produce less diversified results, thus impacting", + "bbox": [ + 511, + 704, + 906, + 900 + ], + "page_idx": 6 + }, + { + "type": "page_number", + "text": "18385", + "bbox": [ + 480, + 944, + 517, + 955 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/dd20bf8d00fb9698e3590f4525d03862108af5f8676c16e5fb68b4693fef7b24.jpg", + "table_caption": [ + "Table 3. Ablation study on the VOA loss. ${\\mathcal{L}}_{\\mathrm{v}},{\\mathcal{L}}_{\\mathrm{o}}$ and ${\\mathcal{L}}_{\\mathrm{b}}$ represent the loss on visible, occluded and background regions, respectively." + ], + "table_footnote": [], + "table_body": "
LvLoLbFID ↓Clip-Score ↑LA ↑OA ↑
✓11.4128.0542.1360.50
✓✓10.2329.2546.4261.50
✓✓14.6528.4244.4777.50
✓✓11.6828.1029.1458.00
✓✓✓10.0329.7355.1182.50
", + "bbox": [ + 104, + 121, + 470, + 222 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/8536f1f2a099ef466ac33c162f3050e267cd2dec4c9d3031f865bd85476b664e.jpg", + "image_caption": [ + "Layout+Order", + "\"A man sitting in front of a table, a laptop and a cup on the table\"" + ], + "image_footnote": [], + "bbox": [ + 109, + 246, + 187, + 303 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/04ae46556d5dac17117f5acbc6f8fe93b109eacf9f266bca9425045c7ded64d0.jpg", + "image_caption": [ + "GLIGEN[30] ControlNet[79] MIGC[83] InstanceDiff.[65]", + "Figure 6. Enhancement of existing methods using VODiff. Row 1 and 2 show the vanilla and the enhanced results, respectively." + ], + "image_footnote": [], + "bbox": [ + 196, + 246, + 468, + 349 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "the FID metrics. ⑧ Although constant visual guidance can improve FID, CLIP-Score, and LA, when combined with VOA, it confuses the order of objects in overlapping regions during the denoising process, interfering with attention optimization and worsening the OA. Finally, combining all the proposed components helps achieve the best results.", + "bbox": [ + 88, + 412, + 482, + 503 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Visibility-Order-Aware (VOA) loss. We verify the efficacy of different loss items in the VOA by removing them separately. Tab. 3 shows the quantitative results, and Supp. Fig. 3 shows the visual results. We can see that $\\mathbf{1} \\mathcal{L}_v$ is indispensable for layout and occlusion constraints (4th row), but with $\\mathcal{L}_v$ alone is also insufficient (1st row); $\\mathbf{2} \\mathcal{L}_v + \\mathcal{L}_b$ produces inferior OA results as the overlapping regions lack constraints; $\\mathbf{3} \\mathcal{L}_v + \\mathcal{L}_o$ has the remarkable OA results, but the worst FID because this combination only considers each object by itself but ignores background constraints (e.g., misaligned \"man\" and \"dog\" in the 1st row, 5-th column of Fig. 3 in Supp.). Finally, combining all items helps achieve the best performance on all metrics.", + "bbox": [ + 89, + 505, + 483, + 700 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.5. Diverse Applications", + "text_level": 1, + "bbox": [ + 89, + 710, + 285, + 728 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Our VODiff is training-free and can be easily integrated into many existing layout-to-image methods [30, 65, 79, 83], significantly enhancing the accuracy of occlusion relationship generation in current pretrained models. As shown in Fig. 6, the result from each method without (or with) using our approach is presented in the first (or second) row. Our VODiff effectively assists these training-based methods in addressing issues such as object disappearance (columns 2-4), as well as incorrect positional (column 5) and occlusion (columns 2 and 5) relationships.", + "bbox": [ + 89, + 733, + 482, + 883 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Our VODiff supports customized generation by integrat-", + "bbox": [ + 109, + 885, + 482, + 900 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/9de675883694e7e7c6e6fd7d0049d94fa2cf0b4b7498f0e0a221bb5a6fae2665.jpg", + "image_caption": [ + "(a) Object insertion with accurate visibility orders" + ], + "image_footnote": [], + "bbox": [ + 517, + 94, + 898, + 162 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/15a480cd80dce19f2db67f37c02eb28b9b5290e460103822fc6e5957453cf8e3.jpg", + "image_caption": [ + "(b) Exemplar based image editing with visibility orders", + "Figure 7. Customized generation with VODiff in object insertion (row 1) and exemplar-based editing (row 2)." + ], + "image_footnote": [], + "bbox": [ + 517, + 175, + 898, + 244 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "ing it with existing techniques, e.g., DreamBooth [56]. In Fig. 7(a), we first demonstrate the insertion of newly generated objects into an existing image via Inversion [59], where the inserted objects adhere to the preset visibility order relative to the original objects in the image, while preserving the background unchanged. Fig. 7(b) shows the integration of our method with DreamBooth [56], enabling the generated objects to not only maintain the identity of the input objects but also ensure that the positions and visibility order among multiple objects adhere to the preset arrangement.", + "bbox": [ + 511, + 300, + 906, + 452 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/92bcd44445f22b7da1ba9d6868606ec75fa9a0ddec913bdea472536451d1b417.jpg", + "image_caption": [ + "\"A bedroom with a cabinet behind a plant, and there is also a bed, a flower on the right and a boy in front of a lamp who is feeding a dog, and a glass bottle is in front of him.\"", + "Figure 8. A failure case of VODiff. From left to right are: inputs, results of InstanceDiff. [65] and Our method, respectively." + ], + "image_footnote": [], + "bbox": [ + 535, + 462, + 883, + 559 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5. Conclusion", + "text_level": 1, + "bbox": [ + 511, + 640, + 633, + 656 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In this paper, we have introduced VODiff to address the challenges of generating complex scenes with object occlusions. We first propose a Sequential Denoising Process that performs object generation sequentially from the bottom to the top layer based on the visibility order of objects. We then propose a Visibility-Order-Aware loss to constrain object positions and occlusion relationships by transforming the layout and visibility order into cross-attention map optimization. The results on our curated VOBench show superior performances of our method in controlling object spatial position and occlusion relationships.", + "bbox": [ + 511, + 657, + 905, + 824 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "VODiff does has limitations. For example, as shown in Fig. 8, like existing method [65], our method may struggle to handle scenarios where the number of objects is exceptionally high due to the challenge of balancing the image quality with precise spatial and occlusion relationships.", + "bbox": [ + 511, + 825, + 905, + 901 + ], + "page_idx": 7 + }, + { + "type": "page_number", + "text": "18386", + "bbox": [ + 480, + 944, + 517, + 955 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Acknowledgements", + "text_level": 1, + "bbox": [ + 91, + 90, + 258, + 107 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "This work is supported in part by Ant Group and General Program of National Natural Science Foundation of China (NSFC, No. 6207071897) and a GRF from the Research Grants Council of Hong Kong (Ref.: 11205620).", + "bbox": [ + 89, + 114, + 483, + 176 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "References", + "text_level": 1, + "bbox": [ + 91, + 186, + 187, + 203 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[1] Omri Avrahami, Thomas Hayes, Oran Gafni, Sonal Gupta, Yaniv Taigman, Devi Parikh, Dani Lischinski, Ohad Fried, and Xi Yin. Spatext: Spatio-textual representation for controllable image generation. In CVPR, pages 18370-18380, 2023. 3", + "[2] Yogesh Balaji, Seungjun Nah, Xun Huang, Arash Vahdat, Jiaming Song, Qinsheng Zhang, Karsten Kreis, Miika Aittala, Timo Aila, Samuli Laine, et al. ediff-i: Text-to-image diffusion models with an ensemble of expert denoisers. arXiv:2211.01324, 2022. 3.5", + "[3] Omer Bar-Tal, Lior Yariv, Yaron Lipman, and Tali Dekel. Multidiffusion: fusing diffusion paths for controlled image generation. In ICML, pages 1737-1752, 2023. 2, 3, 7, 5", + "[4] Tim Brooks, Aleksander Holynski, and Alexei A Efros. Instructpix2pix: Learning to follow image editing instructions. In CVPR, pages 18392-18402, 2023. 2", + "[5] Hila Chefer, Yuval Alaluf, Yael Vinker, Lior Wolf, and Daniel Cohen-Or. Attend-and-excite: Attention-based semantic guidance for text-to-image diffusion models. ACM TOG, 42(4):1-10, 2023. 3, 1", + "[6] Hong Chen, Yipeng Zhang, Xin Wang, Xuguang Duan, Yuwei Zhou, and Wenwu Zhu. Disenbooth: Disentangled parameter-efficient tuning for subject-driven text-to-image generation. arXiv:2305.03374, 3, 2023. 1", + "[7] Kai Chen, Enze Xie, Zhe Chen, Yibo Wang, HONG Lanqing, Zhenguo Li, and Dit-Yan Yeung. Geodiffusion: Text-prompted geometric control for object detection data generation. In ICLR, 2024. 3", + "[8] Minghao Chen, Iro Laina, and Andrea Vedaldi. Training-free layout control with cross-attention guidance. In WACV, pages 5343-5353, 2024. 3, 2", + "[9] Xi Chen, Lianghua Huang, Yu Liu, Yujun Shen, Deli Zhao, and Hengshuang Zhao. Anydoor: Zero-shot object-level image customization. In CVPR, pages 6593-6602, 2024. 3, 1, 6", + "[10] Jiaxin Cheng, Xiao Liang, Xingjian Shi, Tong He, Tianjun Xiao, and Mu Li. Layoutdiffuse: Adapting foundational diffusion models for layout-to-image generation. arXiv preprint arXiv:2302.08908, 2023. 3", + "[11] Patrick Esser, Robin Rombach, and Bjorn Ommer. Taming transformers for high-resolution image synthesis. In CVPR, pages 12873-12883, 2021. 3", + "[12] Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. In ICML, 2024. 3", + "[13] Weixi Feng, Xuehai He, Tsu-Jui Fu, Varun Jampani, Arjun Akula, Pradyumna Narayana, Sugato Basu, Xin Eric" + ], + "bbox": [ + 93, + 212, + 483, + 900 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "Wang, and William Yang Wang. Training-free structured diffusion guidance for compositional text-to-image synthesis. arXiv:2212.05032, 2022. 3", + "[14] Rinon Gal, Yuval Alaluf, Yuval Atzmon, Or Patashnik, Amit H Bermano, Gal Chechik, and Daniel Cohen-Or. An image is worth one word: Personalizing text-to-image generation using textual inversion. arXiv:2208.01618, 2022. 1", + "[15] Biao Gong, Siteng Huang, Yutong Feng, Shiwei Zhang, Yuyuan Li, and Yu Liu. Check locate rectify: A training-free layout calibration system for text-to-image generation. In CVPR, pages 6624-6634, 2024. 3", + "[16] Shuyang Gu, Dong Chen, Jianmin Bao, Fang Wen, Bo Zhang, Dongdong Chen, Lu Yuan, and Baining Guo. Vector quantized diffusion model for text-to-image synthesis. In CVPR, pages 10696-10706, 2022. 3", + "[17] Amir Hertz, Ron Mokady, Jay Tenenbaum, Kfir Aberman, Yael Pritch, and Daniel Cohen-or. Prompt-to-prompt image editing with cross-attention control. In ICLR, 2023. 5", + "[18] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In NIPS, 2017. 6", + "[19] Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. In NIPS Workshop, 2021. 5", + "[20] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In NIPS, 2020. 2, 3", + "[21] Jonathan Ho, Chitwan Sahara, William Chan, David J Fleet, Mohammad Norouzi, and Tim Salimans. Cascaded diffusion models for high fidelity image generation. JMLR, 23(47):1-33, 2022. 3", + "[22] Hexiang Hu, Kelvin CK Chan, Yu-Chuan Su, Wenhu Chen, Yandong Li, Kihyuk Sohn, Yang Zhao, Xue Ben, Boqing Gong, William Cohen, et al. Instruct-imagen: Image generation with multi-modal instruction. In CVPR, pages 4754-4763, 2024. 3", + "[23] Chengyou Jia, Minnan Luo, Zhuohang Dang, Guang Dai, Xiaojun Chang, Mengmeng Wang, and Jingdong Wang. Ssmg: Spatial-semantic map guided diffusion model for free-form layout-to-image generation. In AAAI, pages 2480–2488, 2024. 3", + "[24] Yuming Jiang, Nanxuan Zhao, Qing Liu, Krishna Kumar Singh, Shuai Yang, Chen Change Loy, and Ziwei Liu. Groupdiff: Diffusion-based group portrait editing. arXiv:2409.14379, 2024. 1", + "[25] Zeyinzi Jiang, Chaojie Mao, Yulin Pan, Zhen Han, and Jingfeng Zhang. Scedit: Efficient and controllable image diffusion generation via skip connection editing. In CVPR, pages 8995–9004, 2024. 3", + "[26] Sungnyun Kim, Junsoo Lee, Kibeom Hong, Daesik Kim, and Namhyuk Ahn. Diffblender: Scalable and composable multimodal text-to-image diffusion models. arXiv:2305.15194, 2023. 3", + "[27] Yunji Kim, Jiyoung Lee, Jin-Hwa Kim, Jung-Woo Ha, and Jun-Yan Zhu. Dense text-to-image generation with attention modulation. In ICCV, pages 7701-7711, 2023. 1, 2, 3, 4, 7, 5" + ], + "bbox": [ + 516, + 92, + 903, + 898 + ], + "page_idx": 8 + }, + { + "type": "page_number", + "text": "18387", + "bbox": [ + 480, + 944, + 517, + 955 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[28] Leheng Li, Weichao Qiu, Xu Yan, Jing He, Kaiqiang Zhou, Yingjie Cai, Qing Lian, Bingbing Liu, and Ying-Cong Chen. Omnibooth: Learning latent control for image synthesis with multi-modal instruction. arXiv:2410.04932, 2024. 1", + "[29] Tianle Li, Max Ku, Cong Wei, and Wenhu Chen. Dreamedit: Subject-driven image editing. arXiv:2306.12624, 2023. 3, 1", + "[30] Yuheng Li, Haotian Liu, Qingyang Wu, Fangzhou Mu, Jianwei Yang, Jianfeng Gao, Chunyuan Li, and Yong Jae Lee. Gligen: Open-set grounded text-to-image generation. In CVPR, pages 22511-22521, 2023. 1, 2, 3, 8, 5, 6, 9", + "[31] Yaqi Li, Han Fang, Zerun Feng, Kaijing Ma, Chao Ban, Xianghao Zang, LanXiang Zhou, Zhongjiang He, Jingyan Chen, Jiani Hu, et al. Goal: Grounded text-to-image synthesis with joint layout alignment tuning. In ACM MM, pages 7055-7064, 2024. 3", + "[32] Chang Liu, Xiangtai Li, and Henghui Ding. Referring image editing: Object-level image editing via referring expressions. In CVPR, pages 13128-13138, 2024. 3, 1", + "[33] Nan Liu, Shuang Li, Yilun Du, Antonio Torralba, and Joshua B Tenenbaum. Compositional visual generation with composable diffusion models. In ECCV, pages 423-439, 2022. 3", + "[34] Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. arXiv:2303.05499, 2023.6.1", + "[35] Xiaoyu Liu, Yuxiang Wei, Ming Liu, Xianhui Lin, Peiran Ren, Xuansong Xie, and Wangmeng Zuo. Smartcontrol: Enhancing controlnet for handling rough visual conditions. arXiv:2404.06451, 2024. 3, 7", + "[36] Yuhao Liu, Zhanghan Ke, Fang Liu, Nanxuan Zhao, and Rynson WH Lau. Diff-plugin: Revitalizing details for diffusion-based low-level tasks. In CVPR, pages 4197-4208, 2024. 3", + "[37] Grace Luo, Trevor Darrell, Oliver Wang, Dan B Goldman, and Aleksander Holynski. Readout guidance: Learning control from diffusion features. In CVPR, pages 8217-8227, 2024. 3", + "[38] Kangfu Mei, Mauricio Delbracio, Hossein Talebi, Zhengzhong Tu, Vishal M Patel, and Peyman Milanfar. Codi: Conditional diffusion distillation for higher-fidelity and faster image generation. In CVPR, pages 9048-9058, 2024. 3", + "[39] Sicheng Mo, Fangzhou Mu, Kuan Heng Lin, Yanli Liu, Bochen Guan, Yin Li, and Bolei Zhou. Freecontrol: Training-free spatial control of any text-to-image diffusion model with any condition. In CVPR, pages 7465-7475, 2024. 3, 4, 6, 7, 5", + "[40] Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In AAAI, pages 4296-4304, 2024. 3", + "[41] Nithin Gopalakrishnan Nair, Jeya Maria Jose Valanarasu, and Vishal M Patel. Maxfusion: Plug&play multi-modal generation in text-to-image diffusion models. In ECCV, pages 93-110, 2025. 3" + ], + "bbox": [ + 91, + 90, + 480, + 900 + ], + "page_idx": 9 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[42] Martin E Newell, RG Newell, and Tom L Sancha. A solution to the hidden surface problem. In Seminal graphics: pioneering efforts that shaped the field, pages 27-32. 1998. 4", + "[43] Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. Glide: Towards photorealistic image generation and editing with text-guided diffusion models. arXiv:2112.10741, 2021. 3", + "[44] NovelAI. Novelai: Ai generated stories and art. 2021. 2", + "[45] Lingzhi Pan, Tong Zhang, Bingyuan Chen, Qi Zhou, Wei Ke, Sabine Süssstrunk, and Mathieu Salzmann. Coherent and multi-modality image inpainting via latent space optimization. arXiv:2407.08019, 2024. 1", + "[46] Yulin Pan, Chaojie Mao, Zeyinzi Jiang, Zhen Han, and Jingfeng Zhang. Locate, assign, refine: Taming customized image inpainting with text-subject guidance. arXiv:2403.19534, 2024. 1", + "[47] Quynh Phung, Songwei Ge, and Jia-Bin Huang. Grounded text-to-image synthesis with attention refocusing. In CVPR, pages 7932-7942, 2024. 2, 3, 4, 6, 7, 5", + "[48] Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv:2307.01952, 2023. 3, 5", + "[49] Zipeng Qi, Guoxi Huang, Chenyang Liu, and Fei Ye. Layered rendering diffusion model for zero-shot guided image synthesis. In ECCV, 2024. 3, 1", + "[50] Can Qin, Shu Zhang, Ning Yu, Yihao Feng, Xinyi Yang, Yingbo Zhou, Huan Wang, Juan Carlos Niebles, Caiming Xiong, Silvio Savarese, et al. Uniconcontrol: A unified diffusion model for controllable visual generation in the wild. In NIPS, 2023. 3", + "[51] Leigang Qu, Shengqiong Wu, Hao Fei, Liqiang Nie, and TatSeng Chua. Layoutllm-t2i: Eliciting layout guidance from llm for text-to-image generation. In ACM MM, pages 643-654, 2023. 3", + "[52] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In ICML, pages 8748-8763, 2021. 3, 4", + "[53] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. JMLR, 21(140):1-67, 2020. 3", + "[54] Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. arXiv:2204.06125, 2022. 2, 3", + "[55] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In CVPR, pages 10684-10695, 2022. 1, 2, 3, 4, 5", + "[56] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine" + ], + "bbox": [ + 514, + 90, + 903, + 900 + ], + "page_idx": 9 + }, + { + "type": "page_number", + "text": "18388", + "bbox": [ + 480, + 945, + 517, + 955 + ], + "page_idx": 9 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "tuning text-to-image diffusion models for subject-driven generation. In CVPR, pages 22500-22510, 2023. 8, 1, 6", + "[57] Chitwan Sahara, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding. In NIPS, pages 36479-36494, 2022. 2, 3", + "[58] Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Wortsman, et al. Laion-5b: An open large-scale dataset for training next generation image-text models. In NIPS, pages 25278–25294, 2022. 3, 1", + "[59] Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. In ICLR, 2021. 2, 3, 8, 1, 6", + "[60] Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In ICLR, 2021. 2, 3", + "[61] Yizhi Song, Zhifei Zhang, Zhe Lin, Scott Cohen, Brian Price, Jianming Zhang, Soo Ye Kim, and Daniel Aliaga. Object-stitch: Object compositing with diffusion model. In CVPR, pages 18310-18319, 2023. 3, 1", + "[62] Yizhi Song, Zhifei Zhang, Zhe Lin, Scott Cohen, Brian Price, Jianming Zhang, Soo Ye Kim, He Zhang, Wei Xiong, and Daniel Aliaga. Imprint: Generative object compositing by learning identity-preserving representation. In CVPR, pages 8048-8058, 2024. 3, 1, 6", + "[63] Yanan Sun, Yanchen Liu, Yinhao Tang, Wenjie Pei, and Kai Chen. Anycontrol: Create your artwork with versatile control on text-to-image generation, 2024. 3, 6, 7", + "[64] Zhenhong Sun, Junyan Wang, Zhiyu Tan, Daoyi Dong, Hailan Ma, Hao Li, and Dong Gong. Eggen: Image generation with multi-entity prior learning through entity guidance. In ACM MM, pages 6637-6645, 2024. 3", + "[65] Xudong Wang, Trevor Darrell, Sai Saketh Rambhatla, Rohit Girdhar, and Ishan Misra. Instancediffusion: Instance-level control for image generation. In CVPR, pages 6232-6242, 2024. 3, 6, 7, 8, 5, 9", + "[66] Yuxiang Wei, Yabo Zhang, Zhilong Ji, Jinfeng Bai, Lei Zhang, and Wangmeng Zuo. Elite: Encoding visual concepts into textual embeddings for customized text-to-image generation. In ICCV, pages 15943-15953, 2023. 1", + "[67] Yinwei Wu, Xianpan Zhou, Bing Ma, Xuefeng Su, Kai Ma, and Xinchao Wang. Ifadapter: Instance feature control for grounded text-to-image generation. arXiv:2409.08240, 2024.3", + "[68] Jiayu Xiao, Henglei Lv, Liang Li, Shuhui Wang, and Qingming Huang. R&b: Region and boundary aware zero-shot grounded text-to-image generation. In ICLR, 2023. 2, 3, 6, 7, 5", + "[69] Shitao Xiao, Yueze Wang, Junjie Zhou, Huaying Yuan, Xingrun Xing, Ruiran Yan, Shuting Wang, Tiejun Huang, and Zheng Liu. Omnigen: Unified image generation. arXiv:2409.11340, 2024. 3" + ], + "bbox": [ + 91, + 92, + 482, + 898 + ], + "page_idx": 10 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[70] Jinheng Xie, Yuexiang Li, Yawen Huang, Haozhe Liu, Wentian Zhang, Yefeng Zheng, and Mike Zheng Shou. Boxdiff: Text-to-image synthesis with training-free box-constrained diffusion. In ICCV, pages 7452-7461, 2023. 2, 3, 6, 7, 5", + "[71] Xingqian Xu, Jiayi Guo, Zhangyang Wang, Gao Huang, Irfan Essa, and Humphrey Shi. Prompt-free diffusion: Taking\" text\" out of text-to-image diffusion models. In CVPR, pages 8682-8692, 2024. 3", + "[72] Binxin Yang, Shuyang Gu, Bo Zhang, Ting Zhang, Xuejin Chen, Xiaoyan Sun, Dong Chen, and Fang Wen. Paint by example: Exemplar-based image editing with diffusion models. In CVPR, pages 18381-18391, 2023. 3, 1, 6", + "[73] Ling Yang, Zhaochen Yu, Chenlin Meng, Minkai Xu, Stefano Ermon, and CUI Bin. Mastering text-to-image diffusion: Recaptioning, planning, and generating with multimodal llms. In ICML, 2024. 3, 6, 7, 5", + "[74] Ling Yang, Zhilong Zhang, Zhaochen Yu, Jingwei Liu, Minkai Xu, Stefano Ermon, and Bin Cui. Cross-modal contextualized diffusion models for text-guided visual generation and editing. arXiv:2402.16627, 2024. 3", + "[75] Yang Yang, Wen Wang, Liang Peng, Chaotian Song, Yao Chen, Hengjia Li, Xiaolong Yang, Qinglin Lu, Deng Cai, Boxi Wu, et al. Lora-composer: Leveraging low-rank adaptation for multi-concept customization in training-free diffusion models. arXiv:2403.11627, 2024. 1", + "[76] Zhengyuan Yang, Jianfeng Wang, Zhe Gan, Linjie Li, Kevin Lin, Chenfei Wu, Nan Duan, Zicheng Liu, Ce Liu, Michael Zeng, et al. Reco: Region-controlled text-to-image generation. In CVPR, pages 14246-14255, 2023. 3", + "[77] Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv:2308.06721, 2023. 1", + "[78] Ziyang Yuan, Mingdeng Cao, Xintao Wang, Zhonggang Qi, Chun Yuan, and Ying Shan. Customnet: Zero-shot object customization with variable-viewpoints in text-to-image diffusion models. arXiv:2310.19784, 2023. 1", + "[79] Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In ICCV, pages 3836-3847, 2023. 3, 6, 7, 8, 5, 9", + "[80] Lirui Zhao, Tianshuo Yang, Wenqi Shao, Yuxin Zhang, Yu Qiao, Ping Luo, Kaipeng Zhang, and Rongrong Ji. Diffree: Text-guided shape free object inpainting with diffusion model. arXiv:2407.16982, 2024. 3, 1", + "[81] Shihao Zhao, Dongdong Chen, Yen-Chun Chen, Jianmin Bao, Shaozhe Hao, Lu Yuan, and Kwan-Yee K Wong. Uni-controlnet: All-in-one control to text-to-image diffusion models. In NIPS, 2023. 3", + "[82] Dewei Zhou, You Li, Fan Ma, Zongxin Yang, and Yi Yang. Mige $^{+ + }$ : Advanced multi-instance generation controller for image synthesis. arXiv:2407.02329, 2024. 3", + "[83] Dewei Zhou, You Li, Fan Ma, Xiaoting Zhang, and Yi Yang. Mige: Multi-instance generation controller for text-to-image synthesis. In CVPR, pages 6818-6828, 2024. 1, 2, 3, 6, 7, 8, 5, 9" + ], + "bbox": [ + 516, + 92, + 905, + 854 + ], + "page_idx": 10 + }, + { + "type": "page_number", + "text": "18389", + "bbox": [ + 480, + 944, + 519, + 955 + ], + "page_idx": 10 + } +] \ No newline at end of file diff --git a/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/3dac837e-c8db-4c3a-a872-8c96ea7194f5_model.json b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/3dac837e-c8db-4c3a-a872-8c96ea7194f5_model.json new file mode 100644 index 0000000000000000000000000000000000000000..18cf888f70e73eb268446303240adbb05db291ff --- /dev/null +++ b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/3dac837e-c8db-4c3a-a872-8c96ea7194f5_model.json @@ -0,0 +1,3456 @@ +[ + [ + { + "type": "header", + "bbox": [ + 0.107, + 0.003, + 0.182, + 0.044 + ], + "angle": 0, + "content": "CVF" + }, + { + "type": "header", + "bbox": [ + 0.237, + 0.001, + 0.812, + 0.047 + ], + "angle": 0, + "content": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore." + }, + { + "type": "title", + "bbox": [ + 0.129, + 0.131, + 0.87, + 0.154 + ], + "angle": 0, + "content": "VODiff: Controlling Object Visibility Order in Text-to-Image Generation" + }, + { + "type": "text", + "bbox": [ + 0.101, + 0.186, + 0.91, + 0.228 + ], + "angle": 0, + "content": "Dong Liang\\(^{1,2}\\) Jinyuan Jia\\(^{1,3,\\dagger}\\) Yuhao Liu\\(^{2,\\dagger}\\) Zhanghan Ke\\(^{2}\\) Hongbo Fu\\(^{4}\\) Rynson W.H. Lau\\(^{2,\\dagger}\\) \n\\(^{1}\\)Tongji University \\(^{2}\\)City University of Hong Kong \\(^{3}\\)HKUST(GZ) \\(^{4}\\)HKUST" + }, + { + "type": "text", + "bbox": [ + 0.15, + 0.231, + 0.853, + 0.263 + ], + "angle": 0, + "content": "sse_liangdong@tongji.edu.cn, jinyuanjia@hkust-gz.edu.cn, yuhaoliu7456@gmail.com, zhanghake2-c@my.cityu.edu.hk, hongbofu@ust.hk, Rynson.Lau@cityu.edu.hk" + }, + { + "type": "image_caption", + "bbox": [ + 0.109, + 0.283, + 0.207, + 0.297 + ], + "angle": 0, + "content": "(a) Layout+Order" + }, + { + "type": "image", + "bbox": [ + 0.097, + 0.3, + 0.22, + 0.395 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.259, + 0.283, + 0.343, + 0.297 + ], + "angle": 0, + "content": "(b) SD 1.5 [55]" + }, + { + "type": "image", + "bbox": [ + 0.237, + 0.301, + 0.359, + 0.395 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.38, + 0.283, + 0.474, + 0.297 + ], + "angle": 0, + "content": "(c) GLIGEN [30]" + }, + { + "type": "image", + "bbox": [ + 0.365, + 0.301, + 0.485, + 0.395 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.49, + 0.283, + 0.619, + 0.297 + ], + "angle": 0, + "content": "(d) DenseDiffusion [27]" + }, + { + "type": "image", + "bbox": [ + 0.489, + 0.301, + 0.616, + 0.395 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.644, + 0.283, + 0.724, + 0.297 + ], + "angle": 0, + "content": "(e) MIGC [83]" + }, + { + "type": "image", + "bbox": [ + 0.618, + 0.301, + 0.745, + 0.395 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.777, + 0.283, + 0.874, + 0.297 + ], + "angle": 0, + "content": "(f) VODiff (Ours)" + }, + { + "type": "image", + "bbox": [ + 0.764, + 0.301, + 0.887, + 0.395 + ], + "angle": 0, + "content": null + }, + { + "type": "image_footnote", + "bbox": [ + 0.28, + 0.397, + 0.731, + 0.412 + ], + "angle": 0, + "content": "\"A picture of a table, with a mug and a cup on it, the mug is on a dish.\"" + }, + { + "type": "image", + "bbox": [ + 0.097, + 0.419, + 0.22, + 0.515 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.237, + 0.42, + 0.358, + 0.514 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.365, + 0.42, + 0.485, + 0.513 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.49, + 0.42, + 0.615, + 0.513 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.618, + 0.42, + 0.745, + 0.514 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.765, + 0.421, + 0.886, + 0.515 + ], + "angle": 0, + "content": null + }, + { + "type": "image_footnote", + "bbox": [ + 0.313, + 0.517, + 0.696, + 0.532 + ], + "angle": 0, + "content": "\"A suitcase with a banana in front of it and a dog behind it.\"" + }, + { + "type": "image", + "bbox": [ + 0.098, + 0.541, + 0.22, + 0.636 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.237, + 0.542, + 0.358, + 0.636 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.364, + 0.542, + 0.483, + 0.636 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.49, + 0.542, + 0.615, + 0.636 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.618, + 0.542, + 0.746, + 0.636 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.764, + 0.542, + 0.886, + 0.637 + ], + "angle": 0, + "content": null + }, + { + "type": "image_footnote", + "bbox": [ + 0.214, + 0.64, + 0.804, + 0.655 + ], + "angle": 0, + "content": "\"A picture of a living room with a table and a plant in front of a sofa, and a dog on the sofa.\"" + }, + { + "type": "image_caption", + "bbox": [ + 0.09, + 0.66, + 0.908, + 0.717 + ], + "angle": 0, + "content": "Figure 1. Existing T2I methods that rely on text prompts (b) and those that combine text and layout conditioning (c-e) often struggle to produce images with accurate occlusion relationships. In this work, we propose a new framework called VODiff, which enhances control over object arrangement by introducing their visibility orders (indicated by numbers above their bounding box in the layout map in (a)) as auxiliary constraints. VODiff enables the generation of images with correct spatial arrangements and occlusion relationships." + }, + { + "type": "title", + "bbox": [ + 0.248, + 0.734, + 0.327, + 0.75 + ], + "angle": 0, + "content": "Abstract" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.768, + 0.483, + 0.874 + ], + "angle": 0, + "content": "Recent advancements in diffusion models have significantly enhanced the performance of text-to-image models in image synthesis. To enable control over the spatial locations of the generated objects, diffusion-based methods typically utilize object layout as an auxiliary input. However, we observe that this approach treats all objects as being on the same layer and neglect their visibility order, lead" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.736, + 0.907, + 0.887 + ], + "angle": 0, + "content": "ing to the synthesis of overlapping objects with incorrect occlusions. To address this limitation, we introduce in this paper a new training-free framework that considers object visibility order explicitly and allows users to place overlapping objects in a stack of layers. Our framework consists of two visibility-based designs. First, we propose a novel Sequential Denoising Process (SDP) to divide the whole image generation into multiple stages for different objects, each stage primarily focuses on an object. Second, we propose a novel Visibility-Order-Aware (VOA) Loss to trans" + }, + { + "type": "page_footnote", + "bbox": [ + 0.116, + 0.887, + 0.274, + 0.901 + ], + "angle": 0, + "content": "†Joint corresponding authors." + }, + { + "type": "page_number", + "bbox": [ + 0.481, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "18379" + } + ], + [ + { + "type": "text", + "bbox": [ + 0.089, + 0.092, + 0.482, + 0.258 + ], + "angle": 0, + "content": "form the layout and occlusion constraints into an attention map optimization process to improve the accuracy of synthesizing object occlusions in complex scenes. By merging these two novel components, our framework, dubbed VODiff, enables the generation of photorealistic images that satisfy user-specified spatial constraints and object occlusion relationships. In addition, we introduce VOBench, a diverse benchmark dataset containing 200 curated samples, each with a reference image, text prompts, object visibility orders and layout maps. We conduct extensive evaluations on this dataset to demonstrate the superiority of our approach." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.287, + 0.223, + 0.303 + ], + "angle": 0, + "content": "1. Introduction" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.312, + 0.483, + 0.403 + ], + "angle": 0, + "content": "With the advancements of diffusion models [20, 59], text-to-image (T2I) models [54, 55, 57] have revolutionized the field of image generation. These large-scale T2I models, e.g., Stable Diffusion [55], can generate diverse, high-quality images conforming to given prompts, facilitating more efficient image editing [4] and artistic creation [44]." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.403, + 0.483, + 0.569 + ], + "angle": 0, + "content": "Despite the success, existing T2I models often struggle to produce satisfactory results in complex scenes involving multiple objects with occlusions. On the one hand, text descriptions alone lack spatial controllability, causing text-based methods to likely generate incorrect object positions and relationships, as shown in Fig. 1(b). On the other hand, although some condition-based methods [27, 30, 83] introduce layout conditions (e.g., bbox) into pre-trained T2I models to improve positional accuracy, they do not consider the occlusion relationship of objects and may produce unsatisfactory results, as shown in Fig. 1(c)-(e)." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.569, + 0.483, + 0.825 + ], + "angle": 0, + "content": "In this work, we observe that when generating a scene with multiple objects, existing methods typically employ a one-layer-for-all strategy to synthesize all objects simultaneously on a single canvas. Such a design requires the model to associate the objects with the input text description and accurately obtain their spatial/occlusion relationships from it, if available. This is inherently challenging and often results in mixed or even incorrect object arrangements. Even when layout constraints [3, 27, 47, 68, 70] are provided, the model may still struggle with synthesizing the correct visibility order of overlapping objects, resulting in inaccurate occlusions. To address this problem, we propose a new training-free Visibility Order Diffusion (VODiff) framework to explicitly model the visibility order of objects by assigning each user-specified object to a separate layer and executing a novel multi-layer denoising process to produce the output image." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.826, + 0.483, + 0.901 + ], + "angle": 0, + "content": "We first propose a novel Sequential Denoising Process (SDP) to render the input objects one by one. SDP stratifies the objects into distinct layers, from bottom to top, according to the visibility order. For example, in the second example of Fig. 1, the layers are ordered with the \"dog\" at the" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.092, + 0.905, + 0.333 + ], + "angle": 0, + "content": "bottom, the \"suitcase\" in the middle, and the \"banana\" at the top. Second, SDP divides the complete denoising process into several sub-stages, with each sub-stage primarily focusing on generating one object. Concurrently, during the denoising process, SDP assigns dynamically varying visual guidance to different objects/layers at different stages using a newly proposed smooth guidance mechanism. To prevent the misalignment of inter-object relations caused by deviations in visual guidance during denoising, we further propose a novel Visibility-Order-Aware (VOA) loss to transform layout and occlusion constraints into attention map optimization. When synthesizing an object, we first divide the corresponding layer into three regions, overlapping, visible, and background regions, based on its visibility w.r.t. those of the other objects, and then apply different constraints on these regions." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.335, + 0.905, + 0.47 + ], + "angle": 0, + "content": "To address the lack of a benchmark in visibility-order-aware T2I generation, we also curate a new benchmark called VOBench, which comprises tailored visibility orders, layouts, referenced images, and prompts. To validate the effectiveness of our VODiff, we conduct extensive evaluations on VOBench. Results show that VODiff generates prompt-consistent objects within the user-specified layout regions and precisely delineates the occlusion relationship, as shown in Fig. 1(f)." + }, + { + "type": "text", + "bbox": [ + 0.533, + 0.472, + 0.833, + 0.485 + ], + "angle": 0, + "content": "Our key contributions can be summarized as:" + }, + { + "type": "text", + "bbox": [ + 0.533, + 0.488, + 0.905, + 0.563 + ], + "angle": 0, + "content": "- To the best of our knowledge, we are the first to consider visibility order when generating images with T2I models. We present a new training-free framework named VODiff, which explicitly models object visibility order to control object occlusion relationships." + }, + { + "type": "text", + "bbox": [ + 0.533, + 0.564, + 0.905, + 0.668 + ], + "angle": 0, + "content": "- We propose two novel designs to support the explicit modeling of object visibility order: a Sequential Denoising Process (SDP) to generate different objects sequentially at various sub-stages of the denoising process, and a Visibility-Order-Aware (VOA) Loss to incorporate occlusion relationships into the attention optimization process." + }, + { + "type": "text", + "bbox": [ + 0.533, + 0.669, + 0.905, + 0.759 + ], + "angle": 0, + "content": "- We curate a visibility-order-aware benchmark called VOBench, and conduct extensive evaluations on it. Results show that compared to existing SOTA methods, VODiff can synthesize more photorealistic images, which satisfy user-specified spatial constraints as well as object visibility order." + }, + { + "type": "list", + "bbox": [ + 0.533, + 0.488, + 0.905, + 0.759 + ], + "angle": 0, + "content": null + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.776, + 0.655, + 0.792 + ], + "angle": 0, + "content": "2. Related Work" + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.802, + 0.761, + 0.818 + ], + "angle": 0, + "content": "2.1. Text-to-Image (T2I) models" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.825, + 0.905, + 0.901 + ], + "angle": 0, + "content": "Substantial progress has been made on T2I models through the use of diffusion models [20, 59, 60]. These models conceptualize image synthesis as an iterative denoising process guided by textual prompts. The denoising is conditioned on textual embeddings generated by language" + }, + { + "type": "page_number", + "bbox": [ + 0.481, + 0.945, + 0.519, + 0.956 + ], + "angle": 0, + "content": "18380" + } + ], + [ + { + "type": "text", + "bbox": [ + 0.09, + 0.092, + 0.482, + 0.198 + ], + "angle": 0, + "content": "encoders [52, 53] and is performed either in pixel space [2, 43, 54, 57] or latent space [12, 16, 36, 48, 55, 74], followed by cascaded super-resolution [21] or latent-to-image decoding [11] for high-resolution image generation. By training on extensive datasets of image-text pairs [58], these models learn a unified generative space in which visual and linguistic modalities are closely associated." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.198, + 0.482, + 0.259 + ], + "angle": 0, + "content": "Nonetheless, current T2I diffusion models still struggle to accurately interpret spatial layout instructions, making it challenging to control the occlusion relationships among multiple objects using textual descriptions alone." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.27, + 0.458, + 0.287 + ], + "angle": 0, + "content": "2.2. Layout-guided Text-to-Image (T2I) Models" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.293, + 0.482, + 0.338 + ], + "angle": 0, + "content": "Layout-guided T2I models aim to generate images that contain objects as specified by the input text description, with their locations aligned with the input layout." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.339, + 0.483, + 0.489 + ], + "angle": 0, + "content": "Training-based Models can be categorized into those that fine-tune pre-trained models [1, 7, 10, 23, 31, 51, 72, 76] using the specific input conditions, and those that train additional plug-ins [25, 30, 35, 37, 38, 40, 50, 67, 69, 79] to support various conditions. Some works achieve instance-level control [64, 65, 82, 83] by assigning individual prompts to each object to precisely control their attributes. However, these methods require additional training on large-scale datasets, resulting in significant computational and labor costs." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.491, + 0.483, + 0.612 + ], + "angle": 0, + "content": "Several methods [22, 26, 71, 81] employ multiple control conditions within a single generation. Sun et al. [63] address the complex relationships among different control conditions by training a multi-control encoder to maintain harmonious results when combining multiple controls. Although these approaches can generate images with harmonious occlusion relationships, they still cannot directly control the occlusion relationship of individual objects." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.613, + 0.483, + 0.748 + ], + "angle": 0, + "content": "Sampling-based Models often integrate multiple denoising processes, with each process targeting a specific component or region of the image. Liu et al. [33] interpret diffusion models as energy-based models and propose two compositional operators to combine the score functions of different conditions in a single sampling step. MultiDiff. [3] and LRDiff. [49] then propose to utilize shared parameters or constraints to ensure consistency/accuracy of each object in the generated image." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.749, + 0.483, + 0.81 + ], + "angle": 0, + "content": "Although several methods [39, 41] have proposed training-free multimodal control for the T2I generation, they primarily focus on enhancing control techniques without specifically addressing the object occlusion issues." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.811, + 0.483, + 0.901 + ], + "angle": 0, + "content": "Attention-based Models can be categorized into forward and backward approaches based on how they employ the attention maps. Forward methods [2, 13, 15, 27, 73] manipulate attention values directly on the feature map. The accuracy of these methods heavily depends on the precision of the layout condition (e.g., segmentation masks). Back-" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.092, + 0.905, + 0.198 + ], + "angle": 0, + "content": "ward methods [5, 8, 47, 68, 70, 73] combine the layout condition with the attention map, and utilize an energy function to translate the spatial constraints into the noisy input for iterative updates. These methods also primarily emphasize on positional accuracy while neglecting the occlusion relationships among objects, leading to the unsatisfactory generation of overlapping regions." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.198, + 0.905, + 0.303 + ], + "angle": 0, + "content": "In addition, while several customized T2I models [9, 29, 32, 61, 62, 72, 80] attempt to integrate specific input objects into designated regions of the generated image by using additional layout input, they overlook the visibility order among the integrated objects and other objects within or near those regions. This limitation hinders their effectiveness in addressing occlusion issues." + }, + { + "type": "title", + "bbox": [ + 0.514, + 0.316, + 0.605, + 0.332 + ], + "angle": 0, + "content": "3. Method" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.342, + 0.905, + 0.508 + ], + "angle": 0, + "content": "In this work, our goal is to enhance the ability of pre-trained text-to-image (T2I) diffusion models to correctly handle occlusion relationships in multi-object scenarios without training. We note that the overlapping regions of the objects are the main challenge, and they affect the occlusion relationship of the objects in the synthesized image. However, it is difficult for T2I models to understand occlusion relations among objects through text prompts only. In addition, current spatial layouts can only help constrain the spatial location of the objects. They do not indicate the order of objects in the overlapping regions." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.508, + 0.905, + 0.719 + ], + "angle": 0, + "content": "To achieve our goal, we propose a training-free method, VODiff, which introduces object visibility order into the layout-guided T2I generation to effectively address the incorrect object arrangement problem. Fig. 2 shows our framework. It consists of two novel visibility-aware designs: a Sequential Denoising Process (SDP) and a Visibility-Order-Aware (VOA) loss. When generating an image, our SDP strategy divides the whole denoising process into several sub-stages according to the number of user-specified objects and generates the specified objects in each stage sequentially in a bottom-up manner. Meanwhile, at each denoising step, our VOA loss applies dynamic losses to regions with different visibility states, ensuring accurate occlusion relationships among the generated objects." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.719, + 0.905, + 0.78 + ], + "angle": 0, + "content": "For the rest of this section, we first briefly introduce the preliminary information of diffusion models and attention layers in Sec. 3.1. We then present how we design the SDP in Sec. 3.2 and detail the VOA loss function in Sec. 3.3." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.789, + 0.655, + 0.803 + ], + "angle": 0, + "content": "3.1. Preliminaries" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.81, + 0.905, + 0.901 + ], + "angle": 0, + "content": "Diffusion Models [20, 59, 60] learns to approximate a data distribution by progressively denoising a random variable sampled from a unit Gaussian distribution \\(\\mathcal{N}(0,I)\\). At a denoising step \\(t\\in [1,\\dots ,T]\\), the forward-time Markov chain is formulated as \\(x_{t} = \\sqrt{1 - \\beta_{t}} x_{t - 1} + \\sqrt{\\beta_{t}}\\epsilon_{t - 1}\\), where \\(\\epsilon \\sim \\mathcal{N}(0,I)\\), and \\(\\beta_{t}\\) is a pre-defined variance schedule." + }, + { + "type": "page_number", + "bbox": [ + 0.481, + 0.945, + 0.518, + 0.956 + ], + "angle": 0, + "content": "18381" + } + ], + [ + { + "type": "image", + "bbox": [ + 0.118, + 0.085, + 0.885, + 0.313 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.089, + 0.322, + 0.908, + 0.392 + ], + "angle": 0, + "content": "Figure 2. Our VODiff incorporates two visibility-aware designs: a sequential denoising process (SDP) and a visibility-order-aware (VOA) loss function. SDP divides the denoising process into several stages according to the visibility order, and performs denoising on different parts sequentially. Meanwhile, the VOA loss is utilized to eliminate the local spatial shifts in the attention layers, further preventing position deviations and incorrect occlusion relationships. Given a text prompt, the spatial layout, and the visibility order of the objects to be synthesized as inputs, our VODiff can generate images that simultaneously adhere to all the constraints." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.401, + 0.484, + 0.492 + ], + "angle": 0, + "content": "The reverse-time denoising process can then be defined as: \\( x_{t - 1} = \\tilde{\\alpha}_t x_t + \\tilde{\\beta}_t \\hat{s}_t + \\sigma_t \\epsilon_t \\), where \\( \\tilde{\\alpha}_t, \\tilde{\\beta}_t \\), and \\( \\sigma_t \\) are calculated coefficients. To generate an image, given a noisy input \\( x_T \\) sampled from \\( \\mathcal{N}(0, I) \\), we run denoiser \\( \\epsilon_\\theta \\) iteratively to obtain intermediate representations \\( x_{T - 1}, \\dots, x_1, x_0 \\) where \\( x_0 \\approx x \\) is the final result." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.493, + 0.484, + 0.743 + ], + "angle": 0, + "content": "Attention Layers are the cornerstone of Stable Diffusion [55]. They used to condition the generation based on the text prompt. A pretrained CLIP encoder [52] is usually utilized to encode the text prompt \\( c = (c_{1}, c_{2}, \\dots, c_{n}) \\), where \\( n \\) is the number of tokens, into text embeddings. These tokens are then used as keys \\( K \\in \\mathbb{R}^{n \\times d} \\) and values \\( V \\in \\mathbb{R}^{n \\times d} \\) in the Cross-Attention layer. The symbol \\( d \\) denotes the feature dimension. For a set of queries \\( Q \\in \\mathbb{R}^{hw \\times d} \\) computed from the feature map of size \\( h \\times w \\), the cross-attention map \\( A_{t} \\) at time step \\( t \\) is calculated as: \\( A_{t} = \\text{softmax}\\left(\\frac{QK^{\\top}}{\\sqrt{d}}\\right) \\in [0, 1]^{hw \\times n} \\), where \\( A_{t} \\) contains \\( n \\) attention maps \\( \\{A_{t}^{1}, \\dots, A_{t}^{n}\\} \\), and \\( A_{t}^{i} \\in [0, 1]^{hw} \\) indicates the strength of association between each word token \\( c_{i} \\) and each spatial location on the feature map. A higher attention value in \\( A_{t}^{i} \\) indicates a higher probability of generating the object \\( c_{i} \\) at that spatial location." + }, + { + "type": "title", + "bbox": [ + 0.09, + 0.757, + 0.402, + 0.773 + ], + "angle": 0, + "content": "3.2. Sequential Denoising Process (SDP)" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.781, + 0.484, + 0.903 + ], + "angle": 0, + "content": "Instead of directly generating all objects on a single layer at once, like what existing works [27, 39, 47] do, we are inspired by the painter's algorithm [42], which sorts all polygons according to their z values and then render them starting from the farthest polygon, and the layer concept in image editing tools such as Photoshop. We propose to divide the generation of the target scene into layers, where each object is assigned a separate layer, and synthesize the ob" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.401, + 0.905, + 0.431 + ], + "angle": 0, + "content": "jects sequentially, layer by layer from bottom to top, with the following steps." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.432, + 0.907, + 0.72 + ], + "angle": 0, + "content": "First, we stratify the target objects into distinct layers according to their visibility order. Specifically, given an input text prompt \\( C \\) containing \\( N \\) target objects; a layout condition represented by bounding boxes \\( B = \\{b_{1}, b_{2}, \\ldots, b_{N}\\} \\), where \\( b_{n} \\) is the bounding box of the \\( n \\)-th target object; and visibility orders \\( V = \\{v_{1}, v_{2}, \\ldots, v_{N}\\} \\), where \\( v_{n} \\) is the visibility order of the \\( n \\)-th target object, we convert the layout condition \\( B \\) into \\( N \\) masks, \\( M = \\{m_{1}, m_{2}, \\ldots, m_{N}\\} \\), where \\( m_{n} \\) is the mask corresponding to the bounding box \\( b_{n} \\) with a value of 1 inside the bounding box region and 0 elsewhere. Meanwhile, we extract \\( N \\) object entities from the input text prompt \\( C \\) and denote the set as \\( \\{c_{1}, c_{2}, \\dots, c_{N}\\} \\). Each bounding box \\( b_{n} \\) is associated with an object prompt \\( c_{n} \\), which describes the content of the target object inside \\( b_{n} \\). We order the \\( N \\) target objects in an increasing object visibility order specified by the user, even though there may be non-overlapping objects. The layers from bottom to top can thus be represented as \\( L = \\{(m_{1}, c_{1}, v_{1}), (m_{2}, c_{2}, v_{2}), \\dots, (m_{N}, c_{N}, v_{N})\\} \\)." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.72, + 0.909, + 0.902 + ], + "angle": 0, + "content": "Second, we divide the reverse-time denoising process (with a total of \\( T \\) steps) of our SDP into two main parts, \\( [T, T_g] \\) and \\( [T_g, 0] \\), where \\( T_g \\) denotes the start time of global denoising. The global denoising is to delineate the overall scene based on the input text prompt \\( C \\). The first part of our SDP focuses on per-layer generation based on layers \\( L \\). We divide it into \\( N \\) sub-stages, \\( S = \\{s_1, s_2, \\dots, s_N\\} \\), starting from the bottom layer (i.e., the object with the lowest visibility order). Each sub-stage \\( s_n \\) contains \\( k \\) denoising steps, ranging from \\( [s_n, s_n + k] \\), i.e., \\( T - T_g = kN \\). Since the diffusion denoising process is reversed, \\( s_n + k \\) represents the start time for substage \\( s_n \\). At each sub-stage, to" + }, + { + "type": "page_number", + "bbox": [ + 0.481, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "18382" + } + ], + [ + { + "type": "text", + "bbox": [ + 0.089, + 0.092, + 0.483, + 0.258 + ], + "angle": 0, + "content": "force \\( c_{n} \\) to be generated within the object region, we explicitly incorporate visual guidance into this region. However, we observe that focusing exclusively on a single object \\( n \\) at each stage can lead to the failure of generating other objects, particularly those whose start time is near \\( T_{g} \\), due to the missing of structure information [2] at later denoising steps. To address this problem, at each stage \\( n \\), we apply visual guidance not only for the target object \\( n \\) but also for other objects \\( j \\) (\\( j \\neq n \\)), to ensure the generated layout adheres to the overall requirements. Our scheme for adding visual guidance can be formulated as:" + }, + { + "type": "equation", + "bbox": [ + 0.204, + 0.264, + 0.483, + 0.28 + ], + "angle": 0, + "content": "\\[\nx _ {t, n} = x _ {t} + \\lambda_ {t, n} \\odot m _ {n}, \\tag {1}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.284, + 0.484, + 0.435 + ], + "angle": 0, + "content": "where \\(\\lambda_{t,n}\\) represents the visual guidance scale of the target object \\(n\\) at time step \\(t\\). Note that we allocate a dynamically varying smooth \\(\\lambda_{t,n}\\) to each object to ensure correct occlusion relationships. Specifically, when processing object \\(n\\) at sub-stage \\(n\\), the SDP assigns a guidance scale \\(a\\) to object \\(n\\) and \\(a/2\\) to all other objects. As the process transitions from sub-stage \\(n\\) to sub-stage \\(n + 1\\), the guidance scale for object \\(n\\) decreases linearly to \\(a/2\\), while the guidance scale for object \\(n + 1\\) increases linearly to \\(a\\), and guidance scale for the other objects remains unchanged, as:" + }, + { + "type": "equation", + "bbox": [ + 0.103, + 0.439, + 0.483, + 0.495 + ], + "angle": 0, + "content": "\\[\n\\lambda_ {t, n} = \\left\\{ \\begin{array}{l l} a - \\frac {a}{2} \\frac {t - s _ {n - 1}}{k} & \\text {i f} s _ {n - 1} < t < = s _ {n - 1} + k, \\\\ \\frac {a}{2} + \\frac {a}{2} \\frac {t - s _ {n}}{k} & \\text {i f} s _ {n} < t < = s _ {n} + k, \\\\ \\frac {a}{2} & \\text {o t h e r w i s e}, \\end{array} \\right. \\tag {2}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.497, + 0.484, + 0.648 + ], + "angle": 0, + "content": "where \\(a\\) denotes the initial visual guidance scale (refer to Sec. D.4 of the Supp. for more discussion) and \\(s_0 = T\\). By assigning a smoothly transitioning visual guidance, we ensure that each sub-stage focuses more on a specific object while maintaining correct occlusion relationships between it and other objects, such that in overlapping regions, objects with a higher visibility order occlude those with a lower one. Finally, we forward the updated noisy input \\(x_{t,n}\\) to the denoising model \\(\\epsilon_{\\theta}\\) to estimate noise for each object separately and then combine them together, as:" + }, + { + "type": "equation", + "bbox": [ + 0.117, + 0.652, + 0.483, + 0.693 + ], + "angle": 0, + "content": "\\[\n\\epsilon_ {\\theta} \\left(x _ {t}, t, C\\right) = \\sum_ {n = 1} ^ {N} \\frac {m _ {n}}{\\sum_ {n = 1} ^ {N} m _ {n}} \\otimes \\epsilon_ {\\theta} \\left(x _ {t, n}, t, c _ {n}\\right), \\tag {3}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.697, + 0.483, + 0.727 + ], + "angle": 0, + "content": "where \\(\\epsilon_{\\theta}(x_t,t,C)\\) is the estimated noise used to update \\(x_{t - 1}\\) via the standard diffusion scheduler." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.736, + 0.4, + 0.753 + ], + "angle": 0, + "content": "3.3. Visibility-Order-Aware (VOA) Loss" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.758, + 0.484, + 0.879 + ], + "angle": 0, + "content": "Although SDP can help constrain target generation within the specified region using visual guidance, the consecutive conv. layers of the denoising model introduce local shifts, which are then amplified by the attention layers. As a result, adding visual guidance at the input level (i.e., \\( x_{t} \\)) as a constraint is insufficient and tends to cause object position deviation and false occlusion relationships (see Fig. 5 column 2 for a visual illustration)." + }, + { + "type": "image", + "bbox": [ + 0.522, + 0.091, + 0.639, + 0.171 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.552, + 0.174, + 0.607, + 0.183 + ], + "angle": 0, + "content": "Layout + Order" + }, + { + "type": "image", + "bbox": [ + 0.652, + 0.091, + 0.77, + 0.173 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.643, + 0.174, + 0.78, + 0.185 + ], + "angle": 0, + "content": "moverl mvisible mbackgrd" + }, + { + "type": "image", + "bbox": [ + 0.782, + 0.091, + 0.898, + 0.172 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.819, + 0.175, + 0.867, + 0.183 + ], + "angle": 0, + "content": "Output Image" + }, + { + "type": "image_caption", + "bbox": [ + 0.513, + 0.195, + 0.905, + 0.251 + ], + "angle": 0, + "content": "Figure 3. Region separation illustration in the VOA loss. \\( m_{n}^{\\mathrm{overlap}} \\) represents the regions where the current object \\( n \\) occludes other objects, \\( m_{n}^{\\mathrm{visible}} \\) and \\( m_{n}^{\\mathrm{background}} \\) denote where the current object should and should not be generated in the final image." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.261, + 0.906, + 0.472 + ], + "angle": 0, + "content": "To address the above problem, we propose constraining the spatial location and occlusion relationship at feature-level. Specifically, when generating an image with multiple target objects, different object prompts usually have different activations in different regions of the cross-attention map. At a given spatial location, the higher the attention value of a prompt, the more likely the corresponding object will be generated in that region [17]. In addition, when two objects overlap each other, their occlusion relationship is directly influenced by their visibility order. Once the order changes, the content of the overlapping region changes accordingly. Based on this, we transform the spatial layout and occlusion relationship among the objects to the cross-attention map and optimize it for correct image generation." + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.473, + 0.906, + 0.752 + ], + "angle": 0, + "content": "As shown in Fig. 3, when processing target object \\( n \\), we first divide the mask \\( m_{n} \\) into three parts, overlapping regions \\( m_{n}^{\\mathrm{overlap}} \\), visible regions \\( m_{n}^{\\mathrm{visible}} \\), and background regions \\( m_{n}^{\\mathrm{background}} \\), based on the visibility order and masks of all objects in the layer set \\( L \\). Specifically, for \\( m_{n}^{\\mathrm{overlap}} \\), we compute the overlapping regions (i.e., dark orange region) of the current object mask \\( m_{n} \\) and the masks of all objects with lower visibility order than the current object (i.e., \\( \\{m_{j}\\} \\), with \\( j \\in [1, n-1] \\)) as \\( m_{n}^{\\mathrm{overlap}} = \\bigcup_{j=1}^{n-1} m_{n} \\odot m_{j} \\). This is because an object located at an upper layer should obtain a higher attention score in the overlapping regions. For \\( m_{n}^{\\mathrm{visible}} \\), we first compute the overlapping regions of the current mask \\( m_{n} \\) with the masks of all objects that have a higher visibility order (i.e., \\( \\{m_{j}\\} \\), with \\( j \\in [n+1, N] \\)). We then subtract these overlapping regions from the current object mask, which can be expressed as \\( m_{n}^{\\mathrm{visible}} = m_{n} - \\bigcup_{j=n+1}^{N} m_{n} \\odot m_{j} \\). \\( m_{n}^{\\mathrm{background}} \\) is then obtained by inverting \\( m_{n}^{\\mathrm{visible}} \\) as \\( m_{n}^{\\mathrm{background}} = 1 - m_{n}^{\\mathrm{visible}} \\)." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.752, + 0.906, + 0.872 + ], + "angle": 0, + "content": "During the denoising phase for the \\(n\\)-th target object at time step \\(t\\), as indicated in Eq. 3, we extract the cross-attention maps referred to by the object prompt \\(c_{n}\\) from each layer and average them to produce \\(\\mathbf{A}_{t,n}\\). We then compute the loss between the attention map \\(\\mathbf{A}_{t,n}\\) and each of the regions. For the overlapping regions, which are affected by the visibility order, we ensure that objects in the upper layers obtain higher attention scores, as:" + }, + { + "type": "equation", + "bbox": [ + 0.573, + 0.884, + 0.905, + 0.904 + ], + "angle": 0, + "content": "\\[\n\\mathcal {L} _ {t, n} ^ {\\mathrm {o}} = \\alpha_ {o} \\cdot \\left[ 1 - \\min \\left(\\mathbf {A} _ {t, n} \\odot m _ {n} ^ {\\text {o v e r l a p}}\\right) \\right]. \\tag {4}\n\\]" + }, + { + "type": "page_footnote", + "bbox": [ + 0.114, + 0.887, + 0.438, + 0.901 + ], + "angle": 0, + "content": "For simplicity, we omit the classifier-free guidance [19] here." + }, + { + "type": "page_number", + "bbox": [ + 0.481, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "18383" + } + ], + [ + { + "type": "header", + "bbox": [ + 0.108, + 0.068, + 0.195, + 0.081 + ], + "angle": 0, + "content": "(a) Layout+Order" + }, + { + "type": "header", + "bbox": [ + 0.202, + 0.068, + 0.504, + 0.083 + ], + "angle": 0, + "content": "(b) InstanceDiff.† [65] (c) AnyControl† [63](d) ControlNet† [79]" + }, + { + "type": "header", + "bbox": [ + 0.517, + 0.068, + 0.596, + 0.082 + ], + "angle": 0, + "content": "(e) MIGC† [83]" + }, + { + "type": "header", + "bbox": [ + 0.61, + 0.069, + 0.697, + 0.082 + ], + "angle": 0, + "content": "(f) RPG-Diff. [73]" + }, + { + "type": "header", + "bbox": [ + 0.708, + 0.069, + 0.804, + 0.082 + ], + "angle": 0, + "content": "(g) FreeControl [39]" + }, + { + "type": "header", + "bbox": [ + 0.83, + 0.069, + 0.871, + 0.081 + ], + "angle": 0, + "content": "(h) Ours" + }, + { + "type": "image", + "bbox": [ + 0.101, + 0.086, + 0.198, + 0.158 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.208, + 0.086, + 0.306, + 0.158 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.308, + 0.086, + 0.401, + 0.158 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.403, + 0.086, + 0.499, + 0.158 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.503, + 0.086, + 0.6, + 0.158 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.605, + 0.086, + 0.701, + 0.158 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.705, + 0.086, + 0.801, + 0.158 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.804, + 0.086, + 0.901, + 0.159 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.11, + 0.161, + 0.308, + 0.174 + ], + "angle": 0, + "content": "\"A picture of a cat, in front of a CD.\"" + }, + { + "type": "image", + "bbox": [ + 0.102, + 0.176, + 0.2, + 0.25 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.209, + 0.175, + 0.306, + 0.248 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.308, + 0.175, + 0.401, + 0.248 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.403, + 0.175, + 0.499, + 0.248 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.503, + 0.175, + 0.599, + 0.248 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.605, + 0.175, + 0.701, + 0.248 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.705, + 0.175, + 0.801, + 0.248 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.804, + 0.175, + 0.901, + 0.248 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.108, + 0.253, + 0.443, + 0.266 + ], + "angle": 0, + "content": "\"A laptop on the table with a plant behind it, and a cup near it.\"" + }, + { + "type": "image", + "bbox": [ + 0.102, + 0.268, + 0.199, + 0.341 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.209, + 0.269, + 0.306, + 0.34 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.308, + 0.269, + 0.401, + 0.34 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.403, + 0.269, + 0.499, + 0.34 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.503, + 0.269, + 0.6, + 0.34 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.605, + 0.269, + 0.701, + 0.34 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.705, + 0.269, + 0.801, + 0.34 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.804, + 0.269, + 0.901, + 0.34 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.107, + 0.345, + 0.486, + 0.358 + ], + "angle": 0, + "content": "\"Left chair and right chair, a lamp and a plant in front of a bookshelf.\"" + }, + { + "type": "image_caption", + "bbox": [ + 0.168, + 0.375, + 0.828, + 0.39 + ], + "angle": 0, + "content": "Figure 4. Qualitative comparison with six SOTA methods in Tab. 1. Training-based methods are labeled with \\(\\dagger\\)." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.402, + 0.484, + 0.464 + ], + "angle": 0, + "content": "We then encourage target object \\( n \\) to be generated within the visible regions by increasing the attention score through \\( \\mathcal{L}_{t,n}^{\\mathrm{v}} \\), and prevent the target deviation by weakening the attention score of the background regions via \\( \\mathcal{L}_{t,n}^{\\mathrm{b}} \\):" + }, + { + "type": "equation", + "bbox": [ + 0.145, + 0.471, + 0.483, + 0.49 + ], + "angle": 0, + "content": "\\[\n\\mathcal {L} _ {t, n} ^ {\\mathrm {v}} = \\alpha_ {v i s} \\cdot \\left[ 1 - \\max \\left(\\mathbf {A} _ {t, n} \\odot m _ {n} ^ {\\text {v i s i b l e}}\\right) \\right], \\tag {5}\n\\]" + }, + { + "type": "equation", + "bbox": [ + 0.156, + 0.492, + 0.483, + 0.511 + ], + "angle": 0, + "content": "\\[\n\\mathcal {L} _ {t, n} ^ {\\mathrm {b}} = \\alpha_ {b a c} \\cdot \\max \\left(\\mathbf {A} _ {t, n} \\odot m _ {n} ^ {\\text {b a c k g r o u n d}}\\right), \\tag {6}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.515, + 0.483, + 0.544 + ], + "angle": 0, + "content": "where \\(\\alpha_{o}\\), \\(\\alpha_{vis}\\) and \\(\\alpha_{bac}\\) are coefficients for the three loss functions. Finally, the total loss is:" + }, + { + "type": "equation", + "bbox": [ + 0.184, + 0.544, + 0.482, + 0.579 + ], + "angle": 0, + "content": "\\[\n\\mathcal {L} _ {t} = \\sum_ {n} ^ {N} \\left(\\mathcal {L} _ {t, n} ^ {\\mathrm {o}} + \\mathcal {L} _ {t, n} ^ {\\mathrm {v}} + \\mathcal {L} _ {t, n} ^ {\\mathrm {b}}\\right). \\tag {7}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.579, + 0.484, + 0.699 + ], + "angle": 0, + "content": "With the visibility-order-aware loss, we modify the noise sample \\( x_{t} \\) at each denoising step to minimize the loss using gradient descent \\( \\hat{x}_{t} \\gets x_{t} - \\alpha \\nabla_{x_{t}}\\mathcal{L}_{t} \\), where \\( \\alpha \\) controls the influence of the optimization in the denoising process. In each step \\( t \\), we iterate the gradient descent process \\( \\tau \\) times. By applying the VOA loss on all objects at each sub-stage, VODiff alleviates the guidance deviation and achieves more accurate object spatial position and occlusions." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.704, + 0.215, + 0.72 + ], + "angle": 0, + "content": "4. Experiment" + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.728, + 0.188, + 0.743 + ], + "angle": 0, + "content": "4.1. Metrics" + }, + { + "type": "text", + "bbox": [ + 0.091, + 0.75, + 0.399, + 0.764 + ], + "angle": 0, + "content": "We adopt a few metrics to evaluate our model:" + }, + { + "type": "text", + "bbox": [ + 0.091, + 0.765, + 0.483, + 0.84 + ], + "angle": 0, + "content": "- FID: We use Fréchet Inception Distance [18] to evaluate the image quality. It compares the distribution of generated images to that of real images. A lower FID score indicates a higher image quality, reflecting closer alignment with real images on visual features and diversity." + }, + { + "type": "text", + "bbox": [ + 0.091, + 0.84, + 0.484, + 0.901 + ], + "angle": 0, + "content": "- CLIP-Score: We use the CLIP-Score to compute the distance between input textual features and generated image features, evaluating the fidelity of the generated image to the textual input." + }, + { + "type": "list", + "bbox": [ + 0.091, + 0.765, + 0.484, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "text", + "bbox": [ + 0.514, + 0.402, + 0.906, + 0.477 + ], + "angle": 0, + "content": "- **Layout Alignment (LA):** We use Grounding-DINO [34] to assess Layout alignment accuracy. It detects bounding boxes for each instance and then compute the mean IoU between the detected and Ground Truth boxes. The box threshold is set to 0.5 to avoid incorrect box detection." + }, + { + "type": "text", + "bbox": [ + 0.514, + 0.478, + 0.906, + 0.6 + ], + "angle": 0, + "content": "- Occlusion Alignment (OA): We employ GPT-4V to determine the occlusion relationships between objects in the generated images by prompting it with the image and the object names contained in the image. We then compare the visual order returned by GPT-4V with the ground truth visibility order. If the two match with each other, we count it as accurate. Finally, we compute the percentage of accurate occlusion relationships from all the samples." + }, + { + "type": "list", + "bbox": [ + 0.514, + 0.402, + 0.906, + 0.6 + ], + "angle": 0, + "content": null + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.608, + 0.625, + 0.623 + ], + "angle": 0, + "content": "4.2.VObench" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.627, + 0.907, + 0.822 + ], + "angle": 0, + "content": "Since there are no visibility-order-aware datasets available for evaluation, we curate a VOBench for this purpose. All the raw image data in VOBench are manually collected from the Internet. The data annotation process is as follows. We first employ GroundingDINO [34] and GPT-4V to generate the bounding boxes and text prompts for each object in the image, and manually assign the visibility order to each object box. Based on the number of boxes detected by GroundingDINO, we then divide the images into four categories containing 2-5 objects. For each category, we manually select 50 images based on the degree of overlap between boxes to make sure that various occlusion relations are in our VOBench." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.833, + 0.882, + 0.849 + ], + "angle": 0, + "content": "4.3. Comparison with State-of-the-Art Methods" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.856, + 0.906, + 0.901 + ], + "angle": 0, + "content": "We compare VODiff with 12 layout-guided text-to-image generation methods, including seven training-free methods: BoxDiff [70], R&B [68], Attention-Refocusing [47]," + }, + { + "type": "page_number", + "bbox": [ + 0.481, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "18384" + } + ], + [ + { + "type": "table_caption", + "bbox": [ + 0.09, + 0.089, + 0.908, + 0.133 + ], + "angle": 0, + "content": "Table 1. Quantitative comparison of our VODiff with training-based and training-free methods. LA and OA represent the layout and occlusion alignment accuracy, respectively. AR denotes the average ranking from the user study, while Q, L, and O indicate that the ranking is based on Quality, Layout, and Occlusion. The best and second-best performances are marked in bold and underlined, respectively." + }, + { + "type": "table", + "bbox": [ + 0.099, + 0.136, + 0.894, + 0.243 + ], + "angle": 0, + "content": "
MetricsTraining-basedTraining-free
SmartCtrl [35]ControlNet [79]MIGC[83]InstanceDiff. [65]AnyControl [63]BoxDiff [70]R&B [68]AR [47]MultiDiff. [3]DenseDiff. [27]FreeCtrl [39]RPG [73]Ours
FID ↓10.5818.919.8210.2110.3521.9912.3714.8716.3421.6817.3511.6610.03
CLIP-Score ↑29.4529.3429.6129.5329.6329.4628.7229.5927.4229.3329.1229.5729.73
LA ↑54.8152.7354.6253.5854.7713.5030.8829.7916.6024.1136.1451.1555.11
OA ↑55.0051.5065.0063.5069.5018.5034.5032.0021.5024.0028.5051.5082.50
AR-Q ↓5.3312.132.476.203.058.1510.054.1312.687.109.005.121.50
AR-L ↓2.166.214.045.113.1212.989.0210.0712.0311.148.087.171.73
AR-O ↓6.255.123.184.022.0512.749.0110.1112.0411.098.037.101.26
" + }, + { + "type": "table_caption", + "bbox": [ + 0.09, + 0.255, + 0.483, + 0.284 + ], + "angle": 0, + "content": "Table 2. Ablation study on components. Constant and Dynamic represent the state of change of guidance scale in the SDP." + }, + { + "type": "table", + "bbox": [ + 0.105, + 0.288, + 0.465, + 0.401 + ], + "angle": 0, + "content": "
SDP \nConstant DynamicVOAFID ↓Clip-Score ↑LA ↑OA ↑
✓9.9827.8532.3354.50
✓13.3429.0143.2573.50
✓✓11.6829.3343.3669.50
✓✓10.0329.7355.1182.50
" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.418, + 0.484, + 0.525 + ], + "angle": 0, + "content": "MultiDiffusion [3], DenseDiffusion [27], FreeControl [39] and RPG-Diffusion [73], and five training-based methods: SmartControl [35], ControlNet [79], AnyControl [63], MIGC [83] and InstanceDiffusion [65]. Since these methods do not support visibility order as an input, we include descriptions of object occlusion relationships in the text prompts to ensure fairness." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.531, + 0.484, + 0.698 + ], + "angle": 0, + "content": "Qualitative Comparison. As shown in Fig. 4, we visually compare VODiff with six top-performing methods. The images generated by VODiff exhibit superior spatial consistency and occlusion handling. Specifically, our method handles varying numbers of objects and degrees of occlusion remarkably well. In contrast, other methods exhibit issues such as incorrect spatial positioning (e.g., the 2nd row, (b-f)) or missing objects (e.g., the 3rd row, (d-f)). In addition, although some methods manage to generate correct spatial relationships, they fail in occlusion handling (e.g., the 1st row, (b,e,f,g))." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.705, + 0.484, + 0.901 + ], + "angle": 0, + "content": "Quantitative Comparison. As shown in Tab. 1, we evaluate our method against both training-free and training-based approaches. Our method significantly outperforms the training-free methods across all metrics. We achieve higher CLIP-Score (29.73), LA (55.11), and OA (82.50) scores, indicating superior image quality, layout alignment, and occlusion handling. Despite being training-free, our method outperforms training-based methods in six out of seven metrics. We achieve higher CLIP-Score (29.73 vs 29.63 by AnyControl), LA (55.11 vs 54.81 by SmartControl), and OA (82.50 vs 69.50 by AnyControl) scores. This demonstrates the effectiveness of our method in generating text-/layout-aligned and occlusion handled images without" + }, + { + "type": "image_caption", + "bbox": [ + 0.529, + 0.259, + 0.602, + 0.271 + ], + "angle": 0, + "content": "Layout + Order" + }, + { + "type": "image", + "bbox": [ + 0.529, + 0.274, + 0.603, + 0.322 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.631, + 0.26, + 0.655, + 0.27 + ], + "angle": 0, + "content": "SDP" + }, + { + "type": "image", + "bbox": [ + 0.606, + 0.272, + 0.674, + 0.323 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.699, + 0.26, + 0.726, + 0.27 + ], + "angle": 0, + "content": "VOA" + }, + { + "type": "image", + "bbox": [ + 0.678, + 0.272, + 0.747, + 0.323 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.757, + 0.26, + 0.813, + 0.27 + ], + "angle": 0, + "content": "VOA+SDPc" + }, + { + "type": "image", + "bbox": [ + 0.75, + 0.272, + 0.819, + 0.323 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.845, + 0.26, + 0.868, + 0.27 + ], + "angle": 0, + "content": "Ours" + }, + { + "type": "image", + "bbox": [ + 0.821, + 0.272, + 0.891, + 0.323 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.538, + 0.324, + 0.753, + 0.335 + ], + "angle": 0, + "content": "\"A bowl of strawberries and a cup on a blanket\"" + }, + { + "type": "image", + "bbox": [ + 0.529, + 0.336, + 0.6, + 0.387 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.606, + 0.336, + 0.674, + 0.387 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.678, + 0.336, + 0.746, + 0.387 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.75, + 0.336, + 0.818, + 0.387 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.822, + 0.336, + 0.89, + 0.387 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.538, + 0.389, + 0.7, + 0.4 + ], + "angle": 0, + "content": "\"A man and a dog in front of a sofa\"" + }, + { + "type": "image_caption", + "bbox": [ + 0.513, + 0.404, + 0.906, + 0.446 + ], + "angle": 0, + "content": "Figure 5. Visual comparison of the ablation study in Tab. 2. \\(\\mathrm{SDP}_c\\) denotes that the constant guidance scale is used. By default, the dynamic guidance scale is used in SDP." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.461, + 0.906, + 0.522 + ], + "angle": 0, + "content": "additional training. Although the training-based method, MIGC, achieves the best FID score of 9.82, our method achieves a similarly FID of 10.03. The slight reduction can be attributed to the optimization on attention maps [47]." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.522, + 0.906, + 0.688 + ], + "angle": 0, + "content": "User Study. We also conduct a user study with 63 participants to assess various methods through subjective evaluation. We present users with ten sets of 8 images along the same input conditions (i.e., prompt and layout map), and ask them to rank the given images based on image quality (AR-Q), layout (AR-L), and occlusion accuracy (AR-O). Each pair of images are shown randomly, and we collect ranking scores from different participants. We then compute the average ranking (AR) for three aspects: quality, layout, and occlusion as shown in Tab. 1. The results show a clear preference for our method among the users." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.69, + 0.667, + 0.705 + ], + "angle": 0, + "content": "4.4. Ablation Study" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.705, + 0.907, + 0.901 + ], + "angle": 0, + "content": "Component Analysis. We ablate our VODiff by removing SDP and VOA separately, and replacing the smooth guidance with constant guidance (denoted as \\(\\mathrm{VOA + SDP}_c\\)). Tab. 2 and Fig. 5 show the quantitative and qualitative results. We have three conclusions. Using only SDP (Dynamic) results in the worst LA, OA accuracy and Clip-Score but produces better FID. This is because SDP only adds visual guidance to guide the denoising direction, which does not impose stronger spatial constraints, and thus does not significantly affect the denoising process. VOA significantly improves the layout and occlusion accuracy by optimizing the cross-attention map, but at the same time, causes the model to produce less diversified results, thus impacting" + }, + { + "type": "page_number", + "bbox": [ + 0.481, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "18385" + } + ], + [ + { + "type": "table_caption", + "bbox": [ + 0.091, + 0.09, + 0.483, + 0.118 + ], + "angle": 0, + "content": "Table 3. Ablation study on the VOA loss. \\( {\\mathcal{L}}_{\\mathrm{v}},{\\mathcal{L}}_{\\mathrm{o}} \\) and \\( {\\mathcal{L}}_{\\mathrm{b}} \\) represent the loss on visible, occluded and background regions, respectively." + }, + { + "type": "table", + "bbox": [ + 0.105, + 0.122, + 0.472, + 0.223 + ], + "angle": 0, + "content": "
LvLoLbFID ↓Clip-Score ↑LA ↑OA ↑
✓11.4128.0542.1360.50
✓✓10.2329.2546.4261.50
✓✓14.6528.4244.4777.50
✓✓11.6828.1029.1458.00
✓✓✓10.0329.7355.1182.50
" + }, + { + "type": "image_caption", + "bbox": [ + 0.114, + 0.234, + 0.184, + 0.245 + ], + "angle": 0, + "content": "Layout+Order" + }, + { + "type": "image_caption", + "bbox": [ + 0.197, + 0.234, + 0.478, + 0.246 + ], + "angle": 0, + "content": "GLIGEN[30] ControlNet[79] MIGC[83] InstanceDiff.[65]" + }, + { + "type": "image", + "bbox": [ + 0.11, + 0.247, + 0.188, + 0.304 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.11, + 0.31, + 0.186, + 0.35 + ], + "angle": 0, + "content": "\"A man sitting in front of a table, a laptop and a cup on the table\"" + }, + { + "type": "image", + "bbox": [ + 0.197, + 0.247, + 0.47, + 0.351 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.09, + 0.362, + 0.483, + 0.391 + ], + "angle": 0, + "content": "Figure 6. Enhancement of existing methods using VODiff. Row 1 and 2 show the vanilla and the enhanced results, respectively." + }, + { + "type": "text", + "bbox": [ + 0.089, + 0.414, + 0.483, + 0.505 + ], + "angle": 0, + "content": "the FID metrics. ⑧ Although constant visual guidance can improve FID, CLIP-Score, and LA, when combined with VOA, it confuses the order of objects in overlapping regions during the denoising process, interfering with attention optimization and worsening the OA. Finally, combining all the proposed components helps achieve the best results." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.506, + 0.484, + 0.702 + ], + "angle": 0, + "content": "Visibility-Order-Aware (VOA) loss. We verify the efficacy of different loss items in the VOA by removing them separately. Tab. 3 shows the quantitative results, and Supp. Fig. 3 shows the visual results. We can see that \\(\\mathbf{1} \\mathcal{L}_v\\) is indispensable for layout and occlusion constraints (4th row), but with \\(\\mathcal{L}_v\\) alone is also insufficient (1st row); \\(\\mathbf{2} \\mathcal{L}_v + \\mathcal{L}_b\\) produces inferior OA results as the overlapping regions lack constraints; \\(\\mathbf{3} \\mathcal{L}_v + \\mathcal{L}_o\\) has the remarkable OA results, but the worst FID because this combination only considers each object by itself but ignores background constraints (e.g., misaligned \"man\" and \"dog\" in the 1st row, 5-th column of Fig. 3 in Supp.). Finally, combining all items helps achieve the best performance on all metrics." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.712, + 0.287, + 0.729 + ], + "angle": 0, + "content": "4.5. Diverse Applications" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.734, + 0.483, + 0.885 + ], + "angle": 0, + "content": "Our VODiff is training-free and can be easily integrated into many existing layout-to-image methods [30, 65, 79, 83], significantly enhancing the accuracy of occlusion relationship generation in current pretrained models. As shown in Fig. 6, the result from each method without (or with) using our approach is presented in the first (or second) row. Our VODiff effectively assists these training-based methods in addressing issues such as object disappearance (columns 2-4), as well as incorrect positional (column 5) and occlusion (columns 2 and 5) relationships." + }, + { + "type": "text", + "bbox": [ + 0.11, + 0.886, + 0.483, + 0.901 + ], + "angle": 0, + "content": "Our VODiff supports customized generation by integrat-" + }, + { + "type": "image", + "bbox": [ + 0.518, + 0.095, + 0.9, + 0.164 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.609, + 0.165, + 0.815, + 0.175 + ], + "angle": 0, + "content": "(a) Object insertion with accurate visibility orders" + }, + { + "type": "image", + "bbox": [ + 0.518, + 0.176, + 0.9, + 0.245 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.597, + 0.247, + 0.825, + 0.258 + ], + "angle": 0, + "content": "(b) Exemplar based image editing with visibility orders" + }, + { + "type": "image_caption", + "bbox": [ + 0.513, + 0.264, + 0.905, + 0.293 + ], + "angle": 0, + "content": "Figure 7. Customized generation with VODiff in object insertion (row 1) and exemplar-based editing (row 2)." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.301, + 0.907, + 0.453 + ], + "angle": 0, + "content": "ing it with existing techniques, e.g., DreamBooth [56]. In Fig. 7(a), we first demonstrate the insertion of newly generated objects into an existing image via Inversion [59], where the inserted objects adhere to the preset visibility order relative to the original objects in the image, while preserving the background unchanged. Fig. 7(b) shows the integration of our method with DreamBooth [56], enabling the generated objects to not only maintain the identity of the input objects but also ensure that the positions and visibility order among multiple objects adhere to the preset arrangement." + }, + { + "type": "image", + "bbox": [ + 0.536, + 0.463, + 0.884, + 0.56 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.545, + 0.558, + 0.878, + 0.59 + ], + "angle": 0, + "content": "\"A bedroom with a cabinet behind a plant, and there is also a bed, a flower on the right and a boy in front of a lamp who is feeding a dog, and a glass bottle is in front of him.\"" + }, + { + "type": "image_caption", + "bbox": [ + 0.513, + 0.601, + 0.905, + 0.63 + ], + "angle": 0, + "content": "Figure 8. A failure case of VODiff. From left to right are: inputs, results of InstanceDiff. [65] and Our method, respectively." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.641, + 0.634, + 0.657 + ], + "angle": 0, + "content": "5. Conclusion" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.659, + 0.906, + 0.825 + ], + "angle": 0, + "content": "In this paper, we have introduced VODiff to address the challenges of generating complex scenes with object occlusions. We first propose a Sequential Denoising Process that performs object generation sequentially from the bottom to the top layer based on the visibility order of objects. We then propose a Visibility-Order-Aware loss to constrain object positions and occlusion relationships by transforming the layout and visibility order into cross-attention map optimization. The results on our curated VOBench show superior performances of our method in controlling object spatial position and occlusion relationships." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.826, + 0.906, + 0.902 + ], + "angle": 0, + "content": "VODiff does has limitations. For example, as shown in Fig. 8, like existing method [65], our method may struggle to handle scenarios where the number of objects is exceptionally high due to the challenge of balancing the image quality with precise spatial and occlusion relationships." + }, + { + "type": "page_number", + "bbox": [ + 0.481, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "18386" + } + ], + [ + { + "type": "title", + "bbox": [ + 0.092, + 0.091, + 0.259, + 0.108 + ], + "angle": 0, + "content": "Acknowledgements" + }, + { + "type": "text", + "bbox": [ + 0.091, + 0.115, + 0.484, + 0.177 + ], + "angle": 0, + "content": "This work is supported in part by Ant Group and General Program of National Natural Science Foundation of China (NSFC, No. 6207071897) and a GRF from the Research Grants Council of Hong Kong (Ref.: 11205620)." + }, + { + "type": "title", + "bbox": [ + 0.092, + 0.188, + 0.188, + 0.204 + ], + "angle": 0, + "content": "References" + }, + { + "type": "ref_text", + "bbox": [ + 0.101, + 0.213, + 0.484, + 0.282 + ], + "angle": 0, + "content": "[1] Omri Avrahami, Thomas Hayes, Oran Gafni, Sonal Gupta, Yaniv Taigman, Devi Parikh, Dani Lischinski, Ohad Fried, and Xi Yin. Spatext: Spatio-textual representation for controllable image generation. In CVPR, pages 18370-18380, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.1, + 0.284, + 0.484, + 0.351 + ], + "angle": 0, + "content": "[2] Yogesh Balaji, Seungjun Nah, Xun Huang, Arash Vahdat, Jiaming Song, Qinsheng Zhang, Karsten Kreis, Miika Aittala, Timo Aila, Samuli Laine, et al. ediff-i: Text-to-image diffusion models with an ensemble of expert denoisers. arXiv:2211.01324, 2022. 3.5" + }, + { + "type": "ref_text", + "bbox": [ + 0.101, + 0.354, + 0.483, + 0.395 + ], + "angle": 0, + "content": "[3] Omer Bar-Tal, Lior Yariv, Yaron Lipman, and Tali Dekel. Multidiffusion: fusing diffusion paths for controlled image generation. In ICML, pages 1737-1752, 2023. 2, 3, 7, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.397, + 0.482, + 0.437 + ], + "angle": 0, + "content": "[4] Tim Brooks, Aleksander Holynski, and Alexei A Efros. Instructpix2pix: Learning to follow image editing instructions. In CVPR, pages 18392-18402, 2023. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.439, + 0.483, + 0.492 + ], + "angle": 0, + "content": "[5] Hila Chefer, Yuval Alaluf, Yael Vinker, Lior Wolf, and Daniel Cohen-Or. Attend-and-excite: Attention-based semantic guidance for text-to-image diffusion models. ACM TOG, 42(4):1-10, 2023. 3, 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.494, + 0.483, + 0.549 + ], + "angle": 0, + "content": "[6] Hong Chen, Yipeng Zhang, Xin Wang, Xuguang Duan, Yuwei Zhou, and Wenwu Zhu. Disenbooth: Disentangled parameter-efficient tuning for subject-driven text-to-image generation. arXiv:2305.03374, 3, 2023. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.551, + 0.482, + 0.605 + ], + "angle": 0, + "content": "[7] Kai Chen, Enze Xie, Zhe Chen, Yibo Wang, HONG Lanqing, Zhenguo Li, and Dit-Yan Yeung. Geodiffusion: Text-prompted geometric control for object detection data generation. In ICLR, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.607, + 0.482, + 0.648 + ], + "angle": 0, + "content": "[8] Minghao Chen, Iro Laina, and Andrea Vedaldi. Training-free layout control with cross-attention guidance. In WACV, pages 5343-5353, 2024. 3, 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.65, + 0.482, + 0.702 + ], + "angle": 0, + "content": "[9] Xi Chen, Lianghua Huang, Yu Liu, Yujun Shen, Deli Zhao, and Hengshuang Zhao. Anydoor: Zero-shot object-level image customization. In CVPR, pages 6593-6602, 2024. 3, 1, 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.705, + 0.483, + 0.76 + ], + "angle": 0, + "content": "[10] Jiaxin Cheng, Xiao Liang, Xingjian Shi, Tong He, Tianjun Xiao, and Mu Li. Layoutdiffuse: Adapting foundational diffusion models for layout-to-image generation. arXiv preprint arXiv:2302.08908, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.761, + 0.482, + 0.803 + ], + "angle": 0, + "content": "[11] Patrick Esser, Robin Rombach, and Bjorn Ommer. Taming transformers for high-resolution image synthesis. In CVPR, pages 12873-12883, 2021. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.804, + 0.482, + 0.871 + ], + "angle": 0, + "content": "[12] Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. In ICML, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.873, + 0.482, + 0.901 + ], + "angle": 0, + "content": "[13] Weixi Feng, Xuehai He, Tsu-Jui Fu, Varun Jampani, Arjun Akula, Pradyumna Narayana, Sugato Basu, Xin Eric" + }, + { + "type": "list", + "bbox": [ + 0.094, + 0.213, + 0.484, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "ref_text", + "bbox": [ + 0.548, + 0.093, + 0.905, + 0.134 + ], + "angle": 0, + "content": "Wang, and William Yang Wang. Training-free structured diffusion guidance for compositional text-to-image synthesis. arXiv:2212.05032, 2022. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.137, + 0.905, + 0.192 + ], + "angle": 0, + "content": "[14] Rinon Gal, Yuval Alaluf, Yuval Atzmon, Or Patashnik, Amit H Bermano, Gal Chechik, and Daniel Cohen-Or. An image is worth one word: Personalizing text-to-image generation using textual inversion. arXiv:2208.01618, 2022. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.194, + 0.905, + 0.249 + ], + "angle": 0, + "content": "[15] Biao Gong, Siteng Huang, Yutong Feng, Shiwei Zhang, Yuyuan Li, and Yu Liu. Check locate rectify: A training-free layout calibration system for text-to-image generation. In CVPR, pages 6624-6634, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.252, + 0.905, + 0.307 + ], + "angle": 0, + "content": "[16] Shuyang Gu, Dong Chen, Jianmin Bao, Fang Wen, Bo Zhang, Dongdong Chen, Lu Yuan, and Baining Guo. Vector quantized diffusion model for text-to-image synthesis. In CVPR, pages 10696-10706, 2022. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.31, + 0.905, + 0.351 + ], + "angle": 0, + "content": "[17] Amir Hertz, Ron Mokady, Jay Tenenbaum, Kfir Aberman, Yael Pritch, and Daniel Cohen-or. Prompt-to-prompt image editing with cross-attention control. In ICLR, 2023. 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.354, + 0.905, + 0.408 + ], + "angle": 0, + "content": "[18] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In NIPS, 2017. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.412, + 0.905, + 0.439 + ], + "angle": 0, + "content": "[19] Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. In NIPS Workshop, 2021. 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.442, + 0.905, + 0.469 + ], + "angle": 0, + "content": "[20] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In NIPS, 2020. 2, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.472, + 0.905, + 0.526 + ], + "angle": 0, + "content": "[21] Jonathan Ho, Chitwan Sahara, William Chan, David J Fleet, Mohammad Norouzi, and Tim Salimans. Cascaded diffusion models for high fidelity image generation. JMLR, 23(47):1-33, 2022. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.529, + 0.905, + 0.597 + ], + "angle": 0, + "content": "[22] Hexiang Hu, Kelvin CK Chan, Yu-Chuan Su, Wenhu Chen, Yandong Li, Kihyuk Sohn, Yang Zhao, Xue Ben, Boqing Gong, William Cohen, et al. Instruct-imagen: Image generation with multi-modal instruction. In CVPR, pages 4754-4763, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.601, + 0.905, + 0.669 + ], + "angle": 0, + "content": "[23] Chengyou Jia, Minnan Luo, Zhuohang Dang, Guang Dai, Xiaojun Chang, Mengmeng Wang, and Jingdong Wang. Ssmg: Spatial-semantic map guided diffusion model for free-form layout-to-image generation. In AAAI, pages 2480–2488, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.673, + 0.905, + 0.727 + ], + "angle": 0, + "content": "[24] Yuming Jiang, Nanxuan Zhao, Qing Liu, Krishna Kumar Singh, Shuai Yang, Chen Change Loy, and Ziwei Liu. Groupdiff: Diffusion-based group portrait editing. arXiv:2409.14379, 2024. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.73, + 0.905, + 0.785 + ], + "angle": 0, + "content": "[25] Zeyinzi Jiang, Chaojie Mao, Yulin Pan, Zhen Han, and Jingfeng Zhang. Scedit: Efficient and controllable image diffusion generation via skip connection editing. In CVPR, pages 8995–9004, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.788, + 0.905, + 0.842 + ], + "angle": 0, + "content": "[26] Sungnyun Kim, Junsoo Lee, Kibeom Hong, Daesik Kim, and Namhyuk Ahn. Diffblender: Scalable and composable multimodal text-to-image diffusion models. arXiv:2305.15194, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.846, + 0.905, + 0.899 + ], + "angle": 0, + "content": "[27] Yunji Kim, Jiyoung Lee, Jin-Hwa Kim, Jung-Woo Ha, and Jun-Yan Zhu. Dense text-to-image generation with attention modulation. In ICCV, pages 7701-7711, 2023. 1, 2, 3, 4, 7, 5" + }, + { + "type": "list", + "bbox": [ + 0.517, + 0.093, + 0.905, + 0.899 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.481, + 0.945, + 0.519, + 0.956 + ], + "angle": 0, + "content": "18387" + } + ], + [ + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.092, + 0.482, + 0.147 + ], + "angle": 0, + "content": "[28] Leheng Li, Weichao Qiu, Xu Yan, Jing He, Kaiqiang Zhou, Yingjie Cai, Qing Lian, Bingbing Liu, and Ying-Cong Chen. Omnibooth: Learning latent control for image synthesis with multi-modal instruction. arXiv:2410.04932, 2024. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.149, + 0.482, + 0.177 + ], + "angle": 0, + "content": "[29] Tianle Li, Max Ku, Cong Wei, and Wenhu Chen. Dreamedit: Subject-driven image editing. arXiv:2306.12624, 2023. 3, 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.179, + 0.482, + 0.234 + ], + "angle": 0, + "content": "[30] Yuheng Li, Haotian Liu, Qingyang Wu, Fangzhou Mu, Jianwei Yang, Jianfeng Gao, Chunyuan Li, and Yong Jae Lee. Gligen: Open-set grounded text-to-image generation. In CVPR, pages 22511-22521, 2023. 1, 2, 3, 8, 5, 6, 9" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.236, + 0.482, + 0.303 + ], + "angle": 0, + "content": "[31] Yaqi Li, Han Fang, Zerun Feng, Kaijing Ma, Chao Ban, Xianghao Zang, LanXiang Zhou, Zhongjiang He, Jingyan Chen, Jiani Hu, et al. Goal: Grounded text-to-image synthesis with joint layout alignment tuning. In ACM MM, pages 7055-7064, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.306, + 0.482, + 0.348 + ], + "angle": 0, + "content": "[32] Chang Liu, Xiangtai Li, and Henghui Ding. Referring image editing: Object-level image editing via referring expressions. In CVPR, pages 13128-13138, 2024. 3, 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.35, + 0.482, + 0.403 + ], + "angle": 0, + "content": "[33] Nan Liu, Shuang Li, Yilun Du, Antonio Torralba, and Joshua B Tenenbaum. Compositional visual generation with composable diffusion models. In ECCV, pages 423-439, 2022. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.406, + 0.482, + 0.474 + ], + "angle": 0, + "content": "[34] Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. arXiv:2303.05499, 2023.6.1" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.477, + 0.482, + 0.532 + ], + "angle": 0, + "content": "[35] Xiaoyu Liu, Yuxiang Wei, Ming Liu, Xianhui Lin, Peiran Ren, Xuansong Xie, and Wangmeng Zuo. Smartcontrol: Enhancing controlnet for handling rough visual conditions. arXiv:2404.06451, 2024. 3, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.534, + 0.482, + 0.587 + ], + "angle": 0, + "content": "[36] Yuhao Liu, Zhanghan Ke, Fang Liu, Nanxuan Zhao, and Rynson WH Lau. Diff-plugin: Revitalizing details for diffusion-based low-level tasks. In CVPR, pages 4197-4208, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.591, + 0.482, + 0.644 + ], + "angle": 0, + "content": "[37] Grace Luo, Trevor Darrell, Oliver Wang, Dan B Goldman, and Aleksander Holynski. Readout guidance: Learning control from diffusion features. In CVPR, pages 8217-8227, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.647, + 0.482, + 0.715 + ], + "angle": 0, + "content": "[38] Kangfu Mei, Mauricio Delbracio, Hossein Talebi, Zhengzhong Tu, Vishal M Patel, and Peyman Milanfar. Codi: Conditional diffusion distillation for higher-fidelity and faster image generation. In CVPR, pages 9048-9058, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.718, + 0.482, + 0.786 + ], + "angle": 0, + "content": "[39] Sicheng Mo, Fangzhou Mu, Kuan Heng Lin, Yanli Liu, Bochen Guan, Yin Li, and Bolei Zhou. Freecontrol: Training-free spatial control of any text-to-image diffusion model with any condition. In CVPR, pages 7465-7475, 2024. 3, 4, 6, 7, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.789, + 0.482, + 0.843 + ], + "angle": 0, + "content": "[40] Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In AAAI, pages 4296-4304, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.846, + 0.482, + 0.901 + ], + "angle": 0, + "content": "[41] Nithin Gopalakrishnan Nair, Jeya Maria Jose Valanarasu, and Vishal M Patel. Maxfusion: Plug&play multi-modal generation in text-to-image diffusion models. In ECCV, pages 93-110, 2025. 3" + }, + { + "type": "list", + "bbox": [ + 0.093, + 0.092, + 0.482, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.092, + 0.905, + 0.146 + ], + "angle": 0, + "content": "[42] Martin E Newell, RG Newell, and Tom L Sancha. A solution to the hidden surface problem. In Seminal graphics: pioneering efforts that shaped the field, pages 27-32. 1998. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.149, + 0.905, + 0.218 + ], + "angle": 0, + "content": "[43] Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. Glide: Towards photorealistic image generation and editing with text-guided diffusion models. arXiv:2112.10741, 2021. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.221, + 0.882, + 0.234 + ], + "angle": 0, + "content": "[44] NovelAI. Novelai: Ai generated stories and art. 2021. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.236, + 0.905, + 0.289 + ], + "angle": 0, + "content": "[45] Lingzhi Pan, Tong Zhang, Bingyuan Chen, Qi Zhou, Wei Ke, Sabine Süssstrunk, and Mathieu Salzmann. Coherent and multi-modality image inpainting via latent space optimization. arXiv:2407.08019, 2024. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.292, + 0.905, + 0.346 + ], + "angle": 0, + "content": "[46] Yulin Pan, Chaojie Mao, Zeyinzi Jiang, Zhen Han, and Jingfeng Zhang. Locate, assign, refine: Taming customized image inpainting with text-subject guidance. arXiv:2403.19534, 2024. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.349, + 0.905, + 0.39 + ], + "angle": 0, + "content": "[47] Quynh Phung, Songwei Ge, and Jia-Bin Huang. Grounded text-to-image synthesis with attention refocusing. In CVPR, pages 7932-7942, 2024. 2, 3, 4, 6, 7, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.392, + 0.905, + 0.459 + ], + "angle": 0, + "content": "[48] Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv:2307.01952, 2023. 3, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.462, + 0.905, + 0.503 + ], + "angle": 0, + "content": "[49] Zipeng Qi, Guoxi Huang, Chenyang Liu, and Fei Ye. Layered rendering diffusion model for zero-shot guided image synthesis. In ECCV, 2024. 3, 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.505, + 0.905, + 0.573 + ], + "angle": 0, + "content": "[50] Can Qin, Shu Zhang, Ning Yu, Yihao Feng, Xinyi Yang, Yingbo Zhou, Huan Wang, Juan Carlos Niebles, Caiming Xiong, Silvio Savarese, et al. Uniconcontrol: A unified diffusion model for controllable visual generation in the wild. In NIPS, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.576, + 0.905, + 0.63 + ], + "angle": 0, + "content": "[51] Leigang Qu, Shengqiong Wu, Hao Fei, Liqiang Nie, and TatSeng Chua. Layoutllm-t2i: Eliciting layout guidance from llm for text-to-image generation. In ACM MM, pages 643-654, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.632, + 0.905, + 0.701 + ], + "angle": 0, + "content": "[52] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In ICML, pages 8748-8763, 2021. 3, 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.703, + 0.905, + 0.77 + ], + "angle": 0, + "content": "[53] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. JMLR, 21(140):1-67, 2020. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.773, + 0.905, + 0.814 + ], + "angle": 0, + "content": "[54] Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. arXiv:2204.06125, 2022. 2, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.817, + 0.905, + 0.87 + ], + "angle": 0, + "content": "[55] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In CVPR, pages 10684-10695, 2022. 1, 2, 3, 4, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.873, + 0.905, + 0.901 + ], + "angle": 0, + "content": "[56] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine" + }, + { + "type": "list", + "bbox": [ + 0.516, + 0.092, + 0.905, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.481, + 0.946, + 0.519, + 0.956 + ], + "angle": 0, + "content": "18388" + } + ], + [ + { + "type": "ref_text", + "bbox": [ + 0.125, + 0.093, + 0.482, + 0.12 + ], + "angle": 0, + "content": "tuning text-to-image diffusion models for subject-driven generation. In CVPR, pages 22500-22510, 2023. 8, 1, 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.122, + 0.483, + 0.205 + ], + "angle": 0, + "content": "[57] Chitwan Sahara, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding. In NIPS, pages 36479-36494, 2022. 2, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.207, + 0.483, + 0.289 + ], + "angle": 0, + "content": "[58] Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Wortsman, et al. Laion-5b: An open large-scale dataset for training next generation image-text models. In NIPS, pages 25278–25294, 2022. 3, 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.291, + 0.483, + 0.332 + ], + "angle": 0, + "content": "[59] Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. In ICLR, 2021. 2, 3, 8, 1, 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.334, + 0.483, + 0.389 + ], + "angle": 0, + "content": "[60] Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In ICLR, 2021. 2, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.391, + 0.483, + 0.446 + ], + "angle": 0, + "content": "[61] Yizhi Song, Zhifei Zhang, Zhe Lin, Scott Cohen, Brian Price, Jianming Zhang, Soo Ye Kim, and Daniel Aliaga. Object-stitch: Object compositing with diffusion model. In CVPR, pages 18310-18319, 2023. 3, 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.448, + 0.483, + 0.516 + ], + "angle": 0, + "content": "[62] Yizhi Song, Zhifei Zhang, Zhe Lin, Scott Cohen, Brian Price, Jianming Zhang, Soo Ye Kim, He Zhang, Wei Xiong, and Daniel Aliaga. Imprint: Generative object compositing by learning identity-preserving representation. In CVPR, pages 8048-8058, 2024. 3, 1, 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.518, + 0.483, + 0.56 + ], + "angle": 0, + "content": "[63] Yanan Sun, Yanchen Liu, Yinhao Tang, Wenjie Pei, and Kai Chen. Anycontrol: Create your artwork with versatile control on text-to-image generation, 2024. 3, 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.562, + 0.483, + 0.617 + ], + "angle": 0, + "content": "[64] Zhenhong Sun, Junyan Wang, Zhiyu Tan, Daoyi Dong, Hailan Ma, Hao Li, and Dong Gong. Eggen: Image generation with multi-entity prior learning through entity guidance. In ACM MM, pages 6637-6645, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.619, + 0.483, + 0.673 + ], + "angle": 0, + "content": "[65] Xudong Wang, Trevor Darrell, Sai Saketh Rambhatla, Rohit Girdhar, and Ishan Misra. Instancediffusion: Instance-level control for image generation. In CVPR, pages 6232-6242, 2024. 3, 6, 7, 8, 5, 9" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.676, + 0.483, + 0.731 + ], + "angle": 0, + "content": "[66] Yuxiang Wei, Yabo Zhang, Zhilong Ji, Jinfeng Bai, Lei Zhang, and Wangmeng Zuo. Elite: Encoding visual concepts into textual embeddings for customized text-to-image generation. In ICCV, pages 15943-15953, 2023. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.733, + 0.483, + 0.786 + ], + "angle": 0, + "content": "[67] Yinwei Wu, Xianpan Zhou, Bing Ma, Xuefeng Su, Kai Ma, and Xinchao Wang. Ifadapter: Instance feature control for grounded text-to-image generation. arXiv:2409.08240, 2024.3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.789, + 0.483, + 0.843 + ], + "angle": 0, + "content": "[68] Jiayu Xiao, Henglei Lv, Liang Li, Shuhui Wang, and Qingming Huang. R&b: Region and boundary aware zero-shot grounded text-to-image generation. In ICLR, 2023. 2, 3, 6, 7, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.846, + 0.483, + 0.9 + ], + "angle": 0, + "content": "[69] Shitao Xiao, Yueze Wang, Junjie Zhou, Huaying Yuan, Xingrun Xing, Ruiran Yan, Shuting Wang, Tiejun Huang, and Zheng Liu. Omnigen: Unified image generation. arXiv:2409.11340, 2024. 3" + }, + { + "type": "list", + "bbox": [ + 0.093, + 0.093, + 0.483, + 0.9 + ], + "angle": 0, + "content": null + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.093, + 0.906, + 0.148 + ], + "angle": 0, + "content": "[70] Jinheng Xie, Yuexiang Li, Yawen Huang, Haozhe Liu, Wentian Zhang, Yefeng Zheng, and Mike Zheng Shou. Boxdiff: Text-to-image synthesis with training-free box-constrained diffusion. In ICCV, pages 7452-7461, 2023. 2, 3, 6, 7, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.15, + 0.906, + 0.204 + ], + "angle": 0, + "content": "[71] Xingqian Xu, Jiayi Guo, Zhangyang Wang, Gao Huang, Irfan Essa, and Humphrey Shi. Prompt-free diffusion: Taking\" text\" out of text-to-image diffusion models. In CVPR, pages 8682-8692, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.206, + 0.906, + 0.261 + ], + "angle": 0, + "content": "[72] Binxin Yang, Shuyang Gu, Bo Zhang, Ting Zhang, Xuejin Chen, Xiaoyan Sun, Dong Chen, and Fang Wen. Paint by example: Exemplar-based image editing with diffusion models. In CVPR, pages 18381-18391, 2023. 3, 1, 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.263, + 0.906, + 0.317 + ], + "angle": 0, + "content": "[73] Ling Yang, Zhaochen Yu, Chenlin Meng, Minkai Xu, Stefano Ermon, and CUI Bin. Mastering text-to-image diffusion: Recaptioning, planning, and generating with multimodal llms. In ICML, 2024. 3, 6, 7, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.319, + 0.906, + 0.374 + ], + "angle": 0, + "content": "[74] Ling Yang, Zhilong Zhang, Zhaochen Yu, Jingwei Liu, Minkai Xu, Stefano Ermon, and Bin Cui. Cross-modal contextualized diffusion models for text-guided visual generation and editing. arXiv:2402.16627, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.376, + 0.906, + 0.444 + ], + "angle": 0, + "content": "[75] Yang Yang, Wen Wang, Liang Peng, Chaotian Song, Yao Chen, Hengjia Li, Xiaolong Yang, Qinglin Lu, Deng Cai, Boxi Wu, et al. Lora-composer: Leveraging low-rank adaptation for multi-concept customization in training-free diffusion models. arXiv:2403.11627, 2024. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.446, + 0.906, + 0.502 + ], + "angle": 0, + "content": "[76] Zhengyuan Yang, Jianfeng Wang, Zhe Gan, Linjie Li, Kevin Lin, Chenfei Wu, Nan Duan, Zicheng Liu, Ce Liu, Michael Zeng, et al. Reco: Region-controlled text-to-image generation. In CVPR, pages 14246-14255, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.504, + 0.906, + 0.544 + ], + "angle": 0, + "content": "[77] Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv:2308.06721, 2023. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.546, + 0.906, + 0.6 + ], + "angle": 0, + "content": "[78] Ziyang Yuan, Mingdeng Cao, Xintao Wang, Zhonggang Qi, Chun Yuan, and Ying Shan. Customnet: Zero-shot object customization with variable-viewpoints in text-to-image diffusion models. arXiv:2310.19784, 2023. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.602, + 0.906, + 0.644 + ], + "angle": 0, + "content": "[79] Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In ICCV, pages 3836-3847, 2023. 3, 6, 7, 8, 5, 9" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.646, + 0.906, + 0.699 + ], + "angle": 0, + "content": "[80] Lirui Zhao, Tianshuo Yang, Wenqi Shao, Yuxin Zhang, Yu Qiao, Ping Luo, Kaipeng Zhang, and Rongrong Ji. Diffree: Text-guided shape free object inpainting with diffusion model. arXiv:2407.16982, 2024. 3, 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.702, + 0.906, + 0.756 + ], + "angle": 0, + "content": "[81] Shihao Zhao, Dongdong Chen, Yen-Chun Chen, Jianmin Bao, Shaozhe Hao, Lu Yuan, and Kwan-Yee K Wong. Uni-controlnet: All-in-one control to text-to-image diffusion models. In NIPS, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.759, + 0.906, + 0.8 + ], + "angle": 0, + "content": "[82] Dewei Zhou, You Li, Fan Ma, Zongxin Yang, and Yi Yang. Mige \\(^{+ + }\\) : Advanced multi-instance generation controller for image synthesis. arXiv:2407.02329, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.802, + 0.906, + 0.856 + ], + "angle": 0, + "content": "[83] Dewei Zhou, You Li, Fan Ma, Xiaoting Zhang, and Yi Yang. Mige: Multi-instance generation controller for text-to-image synthesis. In CVPR, pages 6818-6828, 2024. 1, 2, 3, 6, 7, 8, 5, 9" + }, + { + "type": "list", + "bbox": [ + 0.517, + 0.093, + 0.906, + 0.856 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.481, + 0.945, + 0.52, + 0.957 + ], + "angle": 0, + "content": "18389" + } + ] +] \ No newline at end of file diff --git a/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/3dac837e-c8db-4c3a-a872-8c96ea7194f5_origin.pdf b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/3dac837e-c8db-4c3a-a872-8c96ea7194f5_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..87c0dbd9940fd8d98a67b769bbb8870ee4dbf4a4 --- /dev/null +++ b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/3dac837e-c8db-4c3a-a872-8c96ea7194f5_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50ba22015d009bd54a9e03f4e3e884a4c751b8406520b67b7e77ca9367dcfcb6 +size 6456123 diff --git a/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/full.md b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/full.md new file mode 100644 index 0000000000000000000000000000000000000000..dd4583489a6f4a33b003f8bd6e3ecb4b06e1bad8 --- /dev/null +++ b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/full.md @@ -0,0 +1,459 @@ +# VODiff: Controlling Object Visibility Order in Text-to-Image Generation + +Dong Liang $^{1,2}$ Jinyuan Jia $^{1,3,\dagger}$ Yuhao Liu $^{2,\dagger}$ Zhanghan Ke $^{2}$ Hongbo Fu $^{4}$ Rynson W.H. Lau $^{2,\dagger}$ $^{1}$ Tongji University $^{2}$ City University of Hong Kong $^{3}$ HKUST(GZ) $^{4}$ HKUST + +sse_liangdong@tongji.edu.cn, jinyuanjia@hkust-gz.edu.cn, yuhaoliu7456@gmail.com, zhanghake2-c@my.cityu.edu.hk, hongbofu@ust.hk, Rynson.Lau@cityu.edu.hk + +![](images/605eb3a26f0b7f7ec9bc5363a4f2b3c1fb8f14102a074ea03cac17478070f9e8.jpg) +(a) Layout+Order + +(b) SD 1.5 [55] +![](images/396ae77a63730ad6b504e4a40fbeb46cb8acba828a8beb07b5bc76d08f7ca37b.jpg) +"A picture of a table, with a mug and a cup on it, the mug is on a dish." + +![](images/90d8390c805022ce98b0af3bd8a069e40e530d985c5c3495f8e28953b43c651b.jpg) +(c) GLIGEN [30] + +![](images/9156d041384b67c029095cc08a48ac276a6775d1663f900496069dd8f81bacb8.jpg) +(d) DenseDiffusion [27] + +![](images/3ce347d46bf265f8084f29a7a91ae70c3642042c6d1324a0846eac8aae94d125.jpg) +(e) MIGC [83] + +![](images/2e1e30381a2348527e12af9930f5f690daeb7d743f2ff622f2309da400d2597f.jpg) +(f) VODiff (Ours) + +![](images/7d2c4545c2adaf30a0ee372b5f280c38b4cfbe7e01ad5f5c7824507005d7b049.jpg) + +![](images/d002f4801e9720ae91a0bb523b77c2a4d91a051492d21f0091802344fc7b52a5.jpg) +"A suitcase with a banana in front of it and a dog behind it." + +![](images/74686a116326c0a5784c05a7b5fd42838d1ce6c0c438688e1adcf37242e2f0e1.jpg) + +![](images/2c8395177c695dd085d36e42b5856418d2e985a9761daf6162d4040bdd94d995.jpg) + +![](images/ad07d422a8159eb26d5d54266d112d32414f2bd332d94e9767163a9e3dc248b6.jpg) + +![](images/0e8847e8ce613ebf337399df08f281f5452a53b69c60edaf0d07a85bf2af47d3.jpg) + +Figure 1. Existing T2I methods that rely on text prompts (b) and those that combine text and layout conditioning (c-e) often struggle to produce images with accurate occlusion relationships. In this work, we propose a new framework called VODiff, which enhances control over object arrangement by introducing their visibility orders (indicated by numbers above their bounding box in the layout map in (a)) as auxiliary constraints. VODiff enables the generation of images with correct spatial arrangements and occlusion relationships. +![](images/e0cd9df4466b5e25d275006dd0334a0aaafe8427b37a82dd13563634a7e6fca4.jpg) +"A picture of a living room with a table and a plant in front of a sofa, and a dog on the sofa." + +![](images/cd961ee444dbf23977023005ab0f2dd6e8abeba45d987059b2b9ffabc9e27a6f.jpg) + +![](images/ba94a955f8739f650d20019a5b7ee45b70244a8f863f81f790cb4325b2158ee3.jpg) + +![](images/7c21e606ccfd361c39ecff52685c2c865629e60a794729a3d1840f630f373b41.jpg) + +![](images/3d4ca8009b08819f411e09f0503f73962549d923b4cc04d31f104db0bc7869c4.jpg) + +![](images/b363900ca272bb5e12e8a111fb2876bbf094d57e1f4087ab4f293a320cf2acb2.jpg) + +# Abstract + +Recent advancements in diffusion models have significantly enhanced the performance of text-to-image models in image synthesis. To enable control over the spatial locations of the generated objects, diffusion-based methods typically utilize object layout as an auxiliary input. However, we observe that this approach treats all objects as being on the same layer and neglect their visibility order, lead + +ing to the synthesis of overlapping objects with incorrect occlusions. To address this limitation, we introduce in this paper a new training-free framework that considers object visibility order explicitly and allows users to place overlapping objects in a stack of layers. Our framework consists of two visibility-based designs. First, we propose a novel Sequential Denoising Process (SDP) to divide the whole image generation into multiple stages for different objects, each stage primarily focuses on an object. Second, we propose a novel Visibility-Order-Aware (VOA) Loss to trans + +form the layout and occlusion constraints into an attention map optimization process to improve the accuracy of synthesizing object occlusions in complex scenes. By merging these two novel components, our framework, dubbed VODiff, enables the generation of photorealistic images that satisfy user-specified spatial constraints and object occlusion relationships. In addition, we introduce VOBench, a diverse benchmark dataset containing 200 curated samples, each with a reference image, text prompts, object visibility orders and layout maps. We conduct extensive evaluations on this dataset to demonstrate the superiority of our approach. + +# 1. Introduction + +With the advancements of diffusion models [20, 59], text-to-image (T2I) models [54, 55, 57] have revolutionized the field of image generation. These large-scale T2I models, e.g., Stable Diffusion [55], can generate diverse, high-quality images conforming to given prompts, facilitating more efficient image editing [4] and artistic creation [44]. + +Despite the success, existing T2I models often struggle to produce satisfactory results in complex scenes involving multiple objects with occlusions. On the one hand, text descriptions alone lack spatial controllability, causing text-based methods to likely generate incorrect object positions and relationships, as shown in Fig. 1(b). On the other hand, although some condition-based methods [27, 30, 83] introduce layout conditions (e.g., bbox) into pre-trained T2I models to improve positional accuracy, they do not consider the occlusion relationship of objects and may produce unsatisfactory results, as shown in Fig. 1(c)-(e). + +In this work, we observe that when generating a scene with multiple objects, existing methods typically employ a one-layer-for-all strategy to synthesize all objects simultaneously on a single canvas. Such a design requires the model to associate the objects with the input text description and accurately obtain their spatial/occlusion relationships from it, if available. This is inherently challenging and often results in mixed or even incorrect object arrangements. Even when layout constraints [3, 27, 47, 68, 70] are provided, the model may still struggle with synthesizing the correct visibility order of overlapping objects, resulting in inaccurate occlusions. To address this problem, we propose a new training-free Visibility Order Diffusion (VODiff) framework to explicitly model the visibility order of objects by assigning each user-specified object to a separate layer and executing a novel multi-layer denoising process to produce the output image. + +We first propose a novel Sequential Denoising Process (SDP) to render the input objects one by one. SDP stratifies the objects into distinct layers, from bottom to top, according to the visibility order. For example, in the second example of Fig. 1, the layers are ordered with the "dog" at the + +bottom, the "suitcase" in the middle, and the "banana" at the top. Second, SDP divides the complete denoising process into several sub-stages, with each sub-stage primarily focusing on generating one object. Concurrently, during the denoising process, SDP assigns dynamically varying visual guidance to different objects/layers at different stages using a newly proposed smooth guidance mechanism. To prevent the misalignment of inter-object relations caused by deviations in visual guidance during denoising, we further propose a novel Visibility-Order-Aware (VOA) loss to transform layout and occlusion constraints into attention map optimization. When synthesizing an object, we first divide the corresponding layer into three regions, overlapping, visible, and background regions, based on its visibility w.r.t. those of the other objects, and then apply different constraints on these regions. + +To address the lack of a benchmark in visibility-order-aware T2I generation, we also curate a new benchmark called VOBench, which comprises tailored visibility orders, layouts, referenced images, and prompts. To validate the effectiveness of our VODiff, we conduct extensive evaluations on VOBench. Results show that VODiff generates prompt-consistent objects within the user-specified layout regions and precisely delineates the occlusion relationship, as shown in Fig. 1(f). + +Our key contributions can be summarized as: + +- To the best of our knowledge, we are the first to consider visibility order when generating images with T2I models. We present a new training-free framework named VODiff, which explicitly models object visibility order to control object occlusion relationships. +- We propose two novel designs to support the explicit modeling of object visibility order: a Sequential Denoising Process (SDP) to generate different objects sequentially at various sub-stages of the denoising process, and a Visibility-Order-Aware (VOA) Loss to incorporate occlusion relationships into the attention optimization process. +- We curate a visibility-order-aware benchmark called VOBench, and conduct extensive evaluations on it. Results show that compared to existing SOTA methods, VODiff can synthesize more photorealistic images, which satisfy user-specified spatial constraints as well as object visibility order. + +# 2. Related Work + +# 2.1. Text-to-Image (T2I) models + +Substantial progress has been made on T2I models through the use of diffusion models [20, 59, 60]. These models conceptualize image synthesis as an iterative denoising process guided by textual prompts. The denoising is conditioned on textual embeddings generated by language + +encoders [52, 53] and is performed either in pixel space [2, 43, 54, 57] or latent space [12, 16, 36, 48, 55, 74], followed by cascaded super-resolution [21] or latent-to-image decoding [11] for high-resolution image generation. By training on extensive datasets of image-text pairs [58], these models learn a unified generative space in which visual and linguistic modalities are closely associated. + +Nonetheless, current T2I diffusion models still struggle to accurately interpret spatial layout instructions, making it challenging to control the occlusion relationships among multiple objects using textual descriptions alone. + +# 2.2. Layout-guided Text-to-Image (T2I) Models + +Layout-guided T2I models aim to generate images that contain objects as specified by the input text description, with their locations aligned with the input layout. + +Training-based Models can be categorized into those that fine-tune pre-trained models [1, 7, 10, 23, 31, 51, 72, 76] using the specific input conditions, and those that train additional plug-ins [25, 30, 35, 37, 38, 40, 50, 67, 69, 79] to support various conditions. Some works achieve instance-level control [64, 65, 82, 83] by assigning individual prompts to each object to precisely control their attributes. However, these methods require additional training on large-scale datasets, resulting in significant computational and labor costs. + +Several methods [22, 26, 71, 81] employ multiple control conditions within a single generation. Sun et al. [63] address the complex relationships among different control conditions by training a multi-control encoder to maintain harmonious results when combining multiple controls. Although these approaches can generate images with harmonious occlusion relationships, they still cannot directly control the occlusion relationship of individual objects. + +Sampling-based Models often integrate multiple denoising processes, with each process targeting a specific component or region of the image. Liu et al. [33] interpret diffusion models as energy-based models and propose two compositional operators to combine the score functions of different conditions in a single sampling step. MultiDiff. [3] and LRDiff. [49] then propose to utilize shared parameters or constraints to ensure consistency/accuracy of each object in the generated image. + +Although several methods [39, 41] have proposed training-free multimodal control for the T2I generation, they primarily focus on enhancing control techniques without specifically addressing the object occlusion issues. + +Attention-based Models can be categorized into forward and backward approaches based on how they employ the attention maps. Forward methods [2, 13, 15, 27, 73] manipulate attention values directly on the feature map. The accuracy of these methods heavily depends on the precision of the layout condition (e.g., segmentation masks). Back- + +ward methods [5, 8, 47, 68, 70, 73] combine the layout condition with the attention map, and utilize an energy function to translate the spatial constraints into the noisy input for iterative updates. These methods also primarily emphasize on positional accuracy while neglecting the occlusion relationships among objects, leading to the unsatisfactory generation of overlapping regions. + +In addition, while several customized T2I models [9, 29, 32, 61, 62, 72, 80] attempt to integrate specific input objects into designated regions of the generated image by using additional layout input, they overlook the visibility order among the integrated objects and other objects within or near those regions. This limitation hinders their effectiveness in addressing occlusion issues. + +# 3. Method + +In this work, our goal is to enhance the ability of pre-trained text-to-image (T2I) diffusion models to correctly handle occlusion relationships in multi-object scenarios without training. We note that the overlapping regions of the objects are the main challenge, and they affect the occlusion relationship of the objects in the synthesized image. However, it is difficult for T2I models to understand occlusion relations among objects through text prompts only. In addition, current spatial layouts can only help constrain the spatial location of the objects. They do not indicate the order of objects in the overlapping regions. + +To achieve our goal, we propose a training-free method, VODiff, which introduces object visibility order into the layout-guided T2I generation to effectively address the incorrect object arrangement problem. Fig. 2 shows our framework. It consists of two novel visibility-aware designs: a Sequential Denoising Process (SDP) and a Visibility-Order-Aware (VOA) loss. When generating an image, our SDP strategy divides the whole denoising process into several sub-stages according to the number of user-specified objects and generates the specified objects in each stage sequentially in a bottom-up manner. Meanwhile, at each denoising step, our VOA loss applies dynamic losses to regions with different visibility states, ensuring accurate occlusion relationships among the generated objects. + +For the rest of this section, we first briefly introduce the preliminary information of diffusion models and attention layers in Sec. 3.1. We then present how we design the SDP in Sec. 3.2 and detail the VOA loss function in Sec. 3.3. + +# 3.1. Preliminaries + +Diffusion Models [20, 59, 60] learns to approximate a data distribution by progressively denoising a random variable sampled from a unit Gaussian distribution $\mathcal{N}(0,I)$ . At a denoising step $t\in [1,\dots ,T]$ , the forward-time Markov chain is formulated as $x_{t} = \sqrt{1 - \beta_{t}} x_{t - 1} + \sqrt{\beta_{t}}\epsilon_{t - 1}$ , where $\epsilon \sim \mathcal{N}(0,I)$ , and $\beta_{t}$ is a pre-defined variance schedule. + +![](images/2e4359b52be4386cfa615893b5865b4fff7bc1f690515d4dacbb3bb7a3af8615.jpg) +Figure 2. Our VODiff incorporates two visibility-aware designs: a sequential denoising process (SDP) and a visibility-order-aware (VOA) loss function. SDP divides the denoising process into several stages according to the visibility order, and performs denoising on different parts sequentially. Meanwhile, the VOA loss is utilized to eliminate the local spatial shifts in the attention layers, further preventing position deviations and incorrect occlusion relationships. Given a text prompt, the spatial layout, and the visibility order of the objects to be synthesized as inputs, our VODiff can generate images that simultaneously adhere to all the constraints. + +The reverse-time denoising process can then be defined as: $x_{t - 1} = \tilde{\alpha}_t x_t + \tilde{\beta}_t \hat{s}_t + \sigma_t \epsilon_t$ , where $\tilde{\alpha}_t, \tilde{\beta}_t$ , and $\sigma_t$ are calculated coefficients. To generate an image, given a noisy input $x_T$ sampled from $\mathcal{N}(0, I)$ , we run denoiser $\epsilon_\theta$ iteratively to obtain intermediate representations $x_{T - 1}, \dots, x_1, x_0$ where $x_0 \approx x$ is the final result. + +Attention Layers are the cornerstone of Stable Diffusion [55]. They used to condition the generation based on the text prompt. A pretrained CLIP encoder [52] is usually utilized to encode the text prompt $c = (c_{1}, c_{2}, \dots, c_{n})$ , where $n$ is the number of tokens, into text embeddings. These tokens are then used as keys $K \in \mathbb{R}^{n \times d}$ and values $V \in \mathbb{R}^{n \times d}$ in the Cross-Attention layer. The symbol $d$ denotes the feature dimension. For a set of queries $Q \in \mathbb{R}^{hw \times d}$ computed from the feature map of size $h \times w$ , the cross-attention map $A_{t}$ at time step $t$ is calculated as: $A_{t} = \text{softmax}\left(\frac{QK^{\top}}{\sqrt{d}}\right) \in [0, 1]^{hw \times n}$ , where $A_{t}$ contains $n$ attention maps $\{A_{t}^{1}, \dots, A_{t}^{n}\}$ , and $A_{t}^{i} \in [0, 1]^{hw}$ indicates the strength of association between each word token $c_{i}$ and each spatial location on the feature map. A higher attention value in $A_{t}^{i}$ indicates a higher probability of generating the object $c_{i}$ at that spatial location. + +# 3.2. Sequential Denoising Process (SDP) + +Instead of directly generating all objects on a single layer at once, like what existing works [27, 39, 47] do, we are inspired by the painter's algorithm [42], which sorts all polygons according to their z values and then render them starting from the farthest polygon, and the layer concept in image editing tools such as Photoshop. We propose to divide the generation of the target scene into layers, where each object is assigned a separate layer, and synthesize the ob + +jects sequentially, layer by layer from bottom to top, with the following steps. + +First, we stratify the target objects into distinct layers according to their visibility order. Specifically, given an input text prompt $C$ containing $N$ target objects; a layout condition represented by bounding boxes $B = \{b_{1}, b_{2}, \ldots, b_{N}\}$ , where $b_{n}$ is the bounding box of the $n$ -th target object; and visibility orders $V = \{v_{1}, v_{2}, \ldots, v_{N}\}$ , where $v_{n}$ is the visibility order of the $n$ -th target object, we convert the layout condition $B$ into $N$ masks, $M = \{m_{1}, m_{2}, \ldots, m_{N}\}$ , where $m_{n}$ is the mask corresponding to the bounding box $b_{n}$ with a value of 1 inside the bounding box region and 0 elsewhere. Meanwhile, we extract $N$ object entities from the input text prompt $C$ and denote the set as $\{c_{1}, c_{2}, \dots, c_{N}\}$ . Each bounding box $b_{n}$ is associated with an object prompt $c_{n}$ , which describes the content of the target object inside $b_{n}$ . We order the $N$ target objects in an increasing object visibility order specified by the user, even though there may be non-overlapping objects. The layers from bottom to top can thus be represented as $L = \{(m_{1}, c_{1}, v_{1}), (m_{2}, c_{2}, v_{2}), \dots, (m_{N}, c_{N}, v_{N})\}$ . + +Second, we divide the reverse-time denoising process (with a total of $T$ steps) of our SDP into two main parts, $[T, T_g]$ and $[T_g, 0]$ , where $T_g$ denotes the start time of global denoising. The global denoising is to delineate the overall scene based on the input text prompt $C$ . The first part of our SDP focuses on per-layer generation based on layers $L$ . We divide it into $N$ sub-stages, $S = \{s_1, s_2, \dots, s_N\}$ , starting from the bottom layer (i.e., the object with the lowest visibility order). Each sub-stage $s_n$ contains $k$ denoising steps, ranging from $[s_n, s_n + k]$ , i.e., $T - T_g = kN$ . Since the diffusion denoising process is reversed, $s_n + k$ represents the start time for substage $s_n$ . At each sub-stage, to + +force $c_{n}$ to be generated within the object region, we explicitly incorporate visual guidance into this region. However, we observe that focusing exclusively on a single object $n$ at each stage can lead to the failure of generating other objects, particularly those whose start time is near $T_{g}$ , due to the missing of structure information [2] at later denoising steps. To address this problem, at each stage $n$ , we apply visual guidance not only for the target object $n$ but also for other objects $j$ ( $j \neq n$ ), to ensure the generated layout adheres to the overall requirements. Our scheme for adding visual guidance can be formulated as: + +$$ +x _ {t, n} = x _ {t} + \lambda_ {t, n} \odot m _ {n}, \tag {1} +$$ + +where $\lambda_{t,n}$ represents the visual guidance scale of the target object $n$ at time step $t$ . Note that we allocate a dynamically varying smooth $\lambda_{t,n}$ to each object to ensure correct occlusion relationships. Specifically, when processing object $n$ at sub-stage $n$ , the SDP assigns a guidance scale $a$ to object $n$ and $a/2$ to all other objects. As the process transitions from sub-stage $n$ to sub-stage $n + 1$ , the guidance scale for object $n$ decreases linearly to $a/2$ , while the guidance scale for object $n + 1$ increases linearly to $a$ , and guidance scale for the other objects remains unchanged, as: + +$$ +\lambda_ {t, n} = \left\{ \begin{array}{l l} a - \frac {a}{2} \frac {t - s _ {n - 1}}{k} & \text {i f} s _ {n - 1} < t < = s _ {n - 1} + k, \\ \frac {a}{2} + \frac {a}{2} \frac {t - s _ {n}}{k} & \text {i f} s _ {n} < t < = s _ {n} + k, \\ \frac {a}{2} & \text {o t h e r w i s e}, \end{array} \right. \tag {2} +$$ + +where $a$ denotes the initial visual guidance scale (refer to Sec. D.4 of the Supp. for more discussion) and $s_0 = T$ . By assigning a smoothly transitioning visual guidance, we ensure that each sub-stage focuses more on a specific object while maintaining correct occlusion relationships between it and other objects, such that in overlapping regions, objects with a higher visibility order occlude those with a lower one. Finally, we forward the updated noisy input $x_{t,n}$ to the denoising model $\epsilon_{\theta}$ to estimate noise for each object separately and then combine them together, as: + +$$ +\epsilon_ {\theta} \left(x _ {t}, t, C\right) = \sum_ {n = 1} ^ {N} \frac {m _ {n}}{\sum_ {n = 1} ^ {N} m _ {n}} \otimes \epsilon_ {\theta} \left(x _ {t, n}, t, c _ {n}\right), \tag {3} +$$ + +where $\epsilon_{\theta}(x_t,t,C)$ is the estimated noise used to update $x_{t - 1}$ via the standard diffusion scheduler. + +# 3.3. Visibility-Order-Aware (VOA) Loss + +Although SDP can help constrain target generation within the specified region using visual guidance, the consecutive conv. layers of the denoising model introduce local shifts, which are then amplified by the attention layers. As a result, adding visual guidance at the input level (i.e., $x_{t}$ ) as a constraint is insufficient and tends to cause object position deviation and false occlusion relationships (see Fig. 5 column 2 for a visual illustration). + +![](images/0726e66f0937cde8976af8f2f45d781c5d3838e665730c3ac944c38a82b01c79.jpg) +Layout + Order + +![](images/b7b3e19830a58b789a6962629bf870cdb82bf3b59d89d78141f629252401ffba.jpg) +moverl mvisible mbackgrd +Figure 3. Region separation illustration in the VOA loss. $m_{n}^{\mathrm{overlap}}$ represents the regions where the current object $n$ occludes other objects, $m_{n}^{\mathrm{visible}}$ and $m_{n}^{\mathrm{background}}$ denote where the current object should and should not be generated in the final image. + +![](images/76a7c050cf5dd8b17b460fa1a69427b6309273834ae26436aa2950603ffac531.jpg) +Output Image + +To address the above problem, we propose constraining the spatial location and occlusion relationship at feature-level. Specifically, when generating an image with multiple target objects, different object prompts usually have different activations in different regions of the cross-attention map. At a given spatial location, the higher the attention value of a prompt, the more likely the corresponding object will be generated in that region [17]. In addition, when two objects overlap each other, their occlusion relationship is directly influenced by their visibility order. Once the order changes, the content of the overlapping region changes accordingly. Based on this, we transform the spatial layout and occlusion relationship among the objects to the cross-attention map and optimize it for correct image generation. + +As shown in Fig. 3, when processing target object $n$ , we first divide the mask $m_{n}$ into three parts, overlapping regions $m_{n}^{\mathrm{overlap}}$ , visible regions $m_{n}^{\mathrm{visible}}$ , and background regions $m_{n}^{\mathrm{background}}$ , based on the visibility order and masks of all objects in the layer set $L$ . Specifically, for $m_{n}^{\mathrm{overlap}}$ , we compute the overlapping regions (i.e., dark orange region) of the current object mask $m_{n}$ and the masks of all objects with lower visibility order than the current object (i.e., $\{m_{j}\}$ , with $j \in [1, n-1]$ ) as $m_{n}^{\mathrm{overlap}} = \bigcup_{j=1}^{n-1} m_{n} \odot m_{j}$ . This is because an object located at an upper layer should obtain a higher attention score in the overlapping regions. For $m_{n}^{\mathrm{visible}}$ , we first compute the overlapping regions of the current mask $m_{n}$ with the masks of all objects that have a higher visibility order (i.e., $\{m_{j}\}$ , with $j \in [n+1, N]$ ). We then subtract these overlapping regions from the current object mask, which can be expressed as $m_{n}^{\mathrm{visible}} = m_{n} - \bigcup_{j=n+1}^{N} m_{n} \odot m_{j}$ . $m_{n}^{\mathrm{background}}$ is then obtained by inverting $m_{n}^{\mathrm{visible}}$ as $m_{n}^{\mathrm{background}} = 1 - m_{n}^{\mathrm{visible}}$ . + +During the denoising phase for the $n$ -th target object at time step $t$ , as indicated in Eq. 3, we extract the cross-attention maps referred to by the object prompt $c_{n}$ from each layer and average them to produce $\mathbf{A}_{t,n}$ . We then compute the loss between the attention map $\mathbf{A}_{t,n}$ and each of the regions. For the overlapping regions, which are affected by the visibility order, we ensure that objects in the upper layers obtain higher attention scores, as: + +$$ +\mathcal {L} _ {t, n} ^ {\mathrm {o}} = \alpha_ {o} \cdot \left[ 1 - \min \left(\mathbf {A} _ {t, n} \odot m _ {n} ^ {\text {o v e r l a p}}\right) \right]. \tag {4} +$$ + +![](images/f286221b7c77d1d015e6b94307e065093d084e920aad7a540bd50ee82e31d13a.jpg) +"A picture of a cat, in front of a CD." + +![](images/ce17ba2ab414c47029b14a0a20f03afb96287a2ec61845bd18ae535b319a9614.jpg) + +![](images/9d734215b5e77186a88ff5776f6b8b030ff6c2c7c48e1b2eb44940b5c2ef9eb3.jpg) + +![](images/ae0960715533bf88598d93aa1a4b70a2e58a5b9d23565941ed7ba18916b0665f.jpg) + +![](images/9df1fea1ee6af4ad2d64997fcbd1a400e5f7a3ca3fb6bd32d23023877c492a3a.jpg) + +![](images/da0711edb9b22e5fdf281266ad5ec27723e5eceeb5cec78dccb2430f09990a8f.jpg) + +![](images/b31f5795205c9628c116d0e5ec8c6ccd4c8ae76d331d965d80864c8ea4538fb7.jpg) + +![](images/ba6d297d1203906f1275aa759b9b5eb5e6079fedfe8cbecfe9d52c6cdab61c4e.jpg) + +![](images/74a4e7daffbb1f5605e43931c895c8613910e365c9a98c008e380c9e4a5ec4f0.jpg) +"A laptop on the table with a plant behind it, and a cup near it." + +![](images/063d3786b086993b97a559f2741de26895aec842663364a2495a8ef5dbe6da10.jpg) + +![](images/563b95716ece1982d690cc18978f54f7343da30aa3588490567b348b7d60b8ed.jpg) + +![](images/0ee7477788993e3024b0a376df272a5d9cef0127487f94255cadb7092d74d8ba.jpg) + +![](images/be4407ad3e94c30cddaea4eda069f97ac82efc83503b3d7b739568a267444da5.jpg) + +![](images/2edd5e00bffc71990355a2151cd9a70192563c824b967ef62ed8d8e5878a8756.jpg) + +![](images/21dd74696ef2ff62afd698794830001a938fa8943442d8ec0f388e11b3a7fcf0.jpg) + +![](images/7e3083d800d017630758cd36e8de6ab22b08836b478e4a84ee7f9aff1ab6c428.jpg) + +![](images/b280df6c33bf337eb85bf78871cc45a18bbc7e12810b012de9d1b3af6746ca54.jpg) +"Left chair and right chair, a lamp and a plant in front of a bookshelf." + +![](images/582497669611b385566a15c4e9b3e0e11549314bd54777fb6575cb48db381b10.jpg) +Figure 4. Qualitative comparison with six SOTA methods in Tab. 1. Training-based methods are labeled with $\dagger$ . + +![](images/657650ba2ba1848ac104b557bea57d507980cc91ad17e2655f20a7914cc106f4.jpg) + +![](images/3d2e2486f491fba8ad002fc48000a214a3012963043f68fa76f9ec785fba5d19.jpg) + +![](images/afeaeed0f2717ad1f2d0133eb34d2c20d8ed2957758e4817b3180e49a6e86e4b.jpg) + +![](images/2613240aa8c6356dd8c6dc2ff6cc69d524b1762a3f90e9205cd64cddc8d06d93.jpg) + +![](images/d91a5301d1caf3d62be6c3decee53a5b22e71be91798fc9d6818783dc72f24ba.jpg) + +![](images/e1c8306413edda32a103e8f9d238359270439bbeb8079cf3253c0cb0a13dbe7e.jpg) + +We then encourage target object $n$ to be generated within the visible regions by increasing the attention score through $\mathcal{L}_{t,n}^{\mathrm{v}}$ , and prevent the target deviation by weakening the attention score of the background regions via $\mathcal{L}_{t,n}^{\mathrm{b}}$ : + +$$ +\mathcal {L} _ {t, n} ^ {\mathrm {v}} = \alpha_ {v i s} \cdot \left[ 1 - \max \left(\mathbf {A} _ {t, n} \odot m _ {n} ^ {\text {v i s i b l e}}\right) \right], \tag {5} +$$ + +$$ +\mathcal {L} _ {t, n} ^ {\mathrm {b}} = \alpha_ {b a c} \cdot \max \left(\mathbf {A} _ {t, n} \odot m _ {n} ^ {\text {b a c k g r o u n d}}\right), \tag {6} +$$ + +where $\alpha_{o}$ , $\alpha_{vis}$ and $\alpha_{bac}$ are coefficients for the three loss functions. Finally, the total loss is: + +$$ +\mathcal {L} _ {t} = \sum_ {n} ^ {N} \left(\mathcal {L} _ {t, n} ^ {\mathrm {o}} + \mathcal {L} _ {t, n} ^ {\mathrm {v}} + \mathcal {L} _ {t, n} ^ {\mathrm {b}}\right). \tag {7} +$$ + +With the visibility-order-aware loss, we modify the noise sample $x_{t}$ at each denoising step to minimize the loss using gradient descent $\hat{x}_{t} \gets x_{t} - \alpha \nabla_{x_{t}}\mathcal{L}_{t}$ , where $\alpha$ controls the influence of the optimization in the denoising process. In each step $t$ , we iterate the gradient descent process $\tau$ times. By applying the VOA loss on all objects at each sub-stage, VODiff alleviates the guidance deviation and achieves more accurate object spatial position and occlusions. + +# 4. Experiment + +# 4.1. Metrics + +We adopt a few metrics to evaluate our model: + +- FID: We use Fréchet Inception Distance [18] to evaluate the image quality. It compares the distribution of generated images to that of real images. A lower FID score indicates a higher image quality, reflecting closer alignment with real images on visual features and diversity. +- CLIP-Score: We use the CLIP-Score to compute the distance between input textual features and generated image features, evaluating the fidelity of the generated image to the textual input. + +- **Layout Alignment (LA):** We use Grounding-DINO [34] to assess Layout alignment accuracy. It detects bounding boxes for each instance and then compute the mean IoU between the detected and Ground Truth boxes. The box threshold is set to 0.5 to avoid incorrect box detection. +- Occlusion Alignment (OA): We employ GPT-4V to determine the occlusion relationships between objects in the generated images by prompting it with the image and the object names contained in the image. We then compare the visual order returned by GPT-4V with the ground truth visibility order. If the two match with each other, we count it as accurate. Finally, we compute the percentage of accurate occlusion relationships from all the samples. + +# 4.2.VObench + +Since there are no visibility-order-aware datasets available for evaluation, we curate a VOBench for this purpose. All the raw image data in VOBench are manually collected from the Internet. The data annotation process is as follows. We first employ GroundingDINO [34] and GPT-4V to generate the bounding boxes and text prompts for each object in the image, and manually assign the visibility order to each object box. Based on the number of boxes detected by GroundingDINO, we then divide the images into four categories containing 2-5 objects. For each category, we manually select 50 images based on the degree of overlap between boxes to make sure that various occlusion relations are in our VOBench. + +# 4.3. Comparison with State-of-the-Art Methods + +We compare VODiff with 12 layout-guided text-to-image generation methods, including seven training-free methods: BoxDiff [70], R&B [68], Attention-Refocusing [47], + +Table 1. Quantitative comparison of our VODiff with training-based and training-free methods. LA and OA represent the layout and occlusion alignment accuracy, respectively. AR denotes the average ranking from the user study, while Q, L, and O indicate that the ranking is based on Quality, Layout, and Occlusion. The best and second-best performances are marked in bold and underlined, respectively. + +
MetricsTraining-basedTraining-free
SmartCtrl [35]ControlNet [79]MIGC[83]InstanceDiff. [65]AnyControl [63]BoxDiff [70]R&B [68]AR [47]MultiDiff. [3]DenseDiff. [27]FreeCtrl [39]RPG [73]Ours
FID ↓10.5818.919.8210.2110.3521.9912.3714.8716.3421.6817.3511.6610.03
CLIP-Score ↑29.4529.3429.6129.5329.6329.4628.7229.5927.4229.3329.1229.5729.73
LA ↑54.8152.7354.6253.5854.7713.5030.8829.7916.6024.1136.1451.1555.11
OA ↑55.0051.5065.0063.5069.5018.5034.5032.0021.5024.0028.5051.5082.50
AR-Q ↓5.3312.132.476.203.058.1510.054.1312.687.109.005.121.50
AR-L ↓2.166.214.045.113.1212.989.0210.0712.0311.148.087.171.73
AR-O ↓6.255.123.184.022.0512.749.0110.1112.0411.098.037.101.26
+ +Table 2. Ablation study on components. Constant and Dynamic represent the state of change of guidance scale in the SDP. + +
SDP +Constant DynamicVOAFID ↓Clip-Score ↑LA ↑OA ↑
✓9.9827.8532.3354.50
✓13.3429.0143.2573.50
✓✓11.6829.3343.3669.50
✓✓10.0329.7355.1182.50
+ +MultiDiffusion [3], DenseDiffusion [27], FreeControl [39] and RPG-Diffusion [73], and five training-based methods: SmartControl [35], ControlNet [79], AnyControl [63], MIGC [83] and InstanceDiffusion [65]. Since these methods do not support visibility order as an input, we include descriptions of object occlusion relationships in the text prompts to ensure fairness. + +Qualitative Comparison. As shown in Fig. 4, we visually compare VODiff with six top-performing methods. The images generated by VODiff exhibit superior spatial consistency and occlusion handling. Specifically, our method handles varying numbers of objects and degrees of occlusion remarkably well. In contrast, other methods exhibit issues such as incorrect spatial positioning (e.g., the 2nd row, (b-f)) or missing objects (e.g., the 3rd row, (d-f)). In addition, although some methods manage to generate correct spatial relationships, they fail in occlusion handling (e.g., the 1st row, (b,e,f,g)). + +Quantitative Comparison. As shown in Tab. 1, we evaluate our method against both training-free and training-based approaches. Our method significantly outperforms the training-free methods across all metrics. We achieve higher CLIP-Score (29.73), LA (55.11), and OA (82.50) scores, indicating superior image quality, layout alignment, and occlusion handling. Despite being training-free, our method outperforms training-based methods in six out of seven metrics. We achieve higher CLIP-Score (29.73 vs 29.63 by AnyControl), LA (55.11 vs 54.81 by SmartControl), and OA (82.50 vs 69.50 by AnyControl) scores. This demonstrates the effectiveness of our method in generating text-/layout-aligned and occlusion handled images without + +![](images/cecbb91620f2c77915d98270bf5fde2d7c24e901ab3f4010a80789567c45d16f.jpg) +"A bowl of strawberries and a cup on a blanket" + +![](images/41c8934317f3065450f9d280f8e83029236964029378cae0582e953881982acd.jpg) +Layout + Order +SDP + +![](images/f8d3e1200333f75876254ebc53169bda58386ae479f89ca832f206c1de7f6068.jpg) +VOA + +![](images/3bac4bcb48d2485fd1ec6048dd06ef7ea760667f5fd8498758ef20b9962b4110.jpg) +VOA+SDPc + +![](images/2e8af02634b88617c21adff2ce9baaeb55674dacddc9f187f1cd5a5d25c3caab.jpg) +Ours + +![](images/03374d504f92570bcf9c0141629f3e33f3305ce8cf55a682808345a9935959b1.jpg) +"A man and a dog in front of a sofa" +Figure 5. Visual comparison of the ablation study in Tab. 2. $\mathrm{SDP}_c$ denotes that the constant guidance scale is used. By default, the dynamic guidance scale is used in SDP. + +![](images/299d32385b9527f212ba2dea35a6140e9074ea35062a227b4059664b609b1b09.jpg) + +![](images/997686e00c4a7784e1cdeb12ce40b00a953f11cf87dcc97cb21ebaf6cf1824d0.jpg) + +![](images/31d3aaf85111125ebeaf052529692cc6c988e95a07f05d51e6f1c22d159fafbc.jpg) + +![](images/8fa3e6cd82c0e860c844c50297115a37476eb73992c9f004202559cbe3ccce5d.jpg) + +additional training. Although the training-based method, MIGC, achieves the best FID score of 9.82, our method achieves a similarly FID of 10.03. The slight reduction can be attributed to the optimization on attention maps [47]. + +User Study. We also conduct a user study with 63 participants to assess various methods through subjective evaluation. We present users with ten sets of 8 images along the same input conditions (i.e., prompt and layout map), and ask them to rank the given images based on image quality (AR-Q), layout (AR-L), and occlusion accuracy (AR-O). Each pair of images are shown randomly, and we collect ranking scores from different participants. We then compute the average ranking (AR) for three aspects: quality, layout, and occlusion as shown in Tab. 1. The results show a clear preference for our method among the users. + +# 4.4. Ablation Study + +Component Analysis. We ablate our VODiff by removing SDP and VOA separately, and replacing the smooth guidance with constant guidance (denoted as $\mathrm{VOA + SDP}_c$ ). Tab. 2 and Fig. 5 show the quantitative and qualitative results. We have three conclusions. Using only SDP (Dynamic) results in the worst LA, OA accuracy and Clip-Score but produces better FID. This is because SDP only adds visual guidance to guide the denoising direction, which does not impose stronger spatial constraints, and thus does not significantly affect the denoising process. VOA significantly improves the layout and occlusion accuracy by optimizing the cross-attention map, but at the same time, causes the model to produce less diversified results, thus impacting + +Table 3. Ablation study on the VOA loss. ${\mathcal{L}}_{\mathrm{v}},{\mathcal{L}}_{\mathrm{o}}$ and ${\mathcal{L}}_{\mathrm{b}}$ represent the loss on visible, occluded and background regions, respectively. + +
LvLoLbFID ↓Clip-Score ↑LA ↑OA ↑
✓11.4128.0542.1360.50
✓✓10.2329.2546.4261.50
✓✓14.6528.4244.4777.50
✓✓11.6828.1029.1458.00
✓✓✓10.0329.7355.1182.50
+ +![](images/8536f1f2a099ef466ac33c162f3050e267cd2dec4c9d3031f865bd85476b664e.jpg) +Layout+Order +"A man sitting in front of a table, a laptop and a cup on the table" + +![](images/04ae46556d5dac17117f5acbc6f8fe93b109eacf9f266bca9425045c7ded64d0.jpg) +GLIGEN[30] ControlNet[79] MIGC[83] InstanceDiff.[65] +Figure 6. Enhancement of existing methods using VODiff. Row 1 and 2 show the vanilla and the enhanced results, respectively. + +the FID metrics. ⑧ Although constant visual guidance can improve FID, CLIP-Score, and LA, when combined with VOA, it confuses the order of objects in overlapping regions during the denoising process, interfering with attention optimization and worsening the OA. Finally, combining all the proposed components helps achieve the best results. + +Visibility-Order-Aware (VOA) loss. We verify the efficacy of different loss items in the VOA by removing them separately. Tab. 3 shows the quantitative results, and Supp. Fig. 3 shows the visual results. We can see that $\mathbf{1} \mathcal{L}_v$ is indispensable for layout and occlusion constraints (4th row), but with $\mathcal{L}_v$ alone is also insufficient (1st row); $\mathbf{2} \mathcal{L}_v + \mathcal{L}_b$ produces inferior OA results as the overlapping regions lack constraints; $\mathbf{3} \mathcal{L}_v + \mathcal{L}_o$ has the remarkable OA results, but the worst FID because this combination only considers each object by itself but ignores background constraints (e.g., misaligned "man" and "dog" in the 1st row, 5-th column of Fig. 3 in Supp.). Finally, combining all items helps achieve the best performance on all metrics. + +# 4.5. Diverse Applications + +Our VODiff is training-free and can be easily integrated into many existing layout-to-image methods [30, 65, 79, 83], significantly enhancing the accuracy of occlusion relationship generation in current pretrained models. As shown in Fig. 6, the result from each method without (or with) using our approach is presented in the first (or second) row. Our VODiff effectively assists these training-based methods in addressing issues such as object disappearance (columns 2-4), as well as incorrect positional (column 5) and occlusion (columns 2 and 5) relationships. + +Our VODiff supports customized generation by integrat- + +![](images/9de675883694e7e7c6e6fd7d0049d94fa2cf0b4b7498f0e0a221bb5a6fae2665.jpg) +(a) Object insertion with accurate visibility orders + +![](images/15a480cd80dce19f2db67f37c02eb28b9b5290e460103822fc6e5957453cf8e3.jpg) +(b) Exemplar based image editing with visibility orders +Figure 7. Customized generation with VODiff in object insertion (row 1) and exemplar-based editing (row 2). + +ing it with existing techniques, e.g., DreamBooth [56]. In Fig. 7(a), we first demonstrate the insertion of newly generated objects into an existing image via Inversion [59], where the inserted objects adhere to the preset visibility order relative to the original objects in the image, while preserving the background unchanged. Fig. 7(b) shows the integration of our method with DreamBooth [56], enabling the generated objects to not only maintain the identity of the input objects but also ensure that the positions and visibility order among multiple objects adhere to the preset arrangement. + +![](images/92bcd44445f22b7da1ba9d6868606ec75fa9a0ddec913bdea472536451d1b417.jpg) +"A bedroom with a cabinet behind a plant, and there is also a bed, a flower on the right and a boy in front of a lamp who is feeding a dog, and a glass bottle is in front of him." +Figure 8. A failure case of VODiff. From left to right are: inputs, results of InstanceDiff. [65] and Our method, respectively. + +# 5. Conclusion + +In this paper, we have introduced VODiff to address the challenges of generating complex scenes with object occlusions. We first propose a Sequential Denoising Process that performs object generation sequentially from the bottom to the top layer based on the visibility order of objects. We then propose a Visibility-Order-Aware loss to constrain object positions and occlusion relationships by transforming the layout and visibility order into cross-attention map optimization. The results on our curated VOBench show superior performances of our method in controlling object spatial position and occlusion relationships. + +VODiff does has limitations. For example, as shown in Fig. 8, like existing method [65], our method may struggle to handle scenarios where the number of objects is exceptionally high due to the challenge of balancing the image quality with precise spatial and occlusion relationships. + +# Acknowledgements + +This work is supported in part by Ant Group and General Program of National Natural Science Foundation of China (NSFC, No. 6207071897) and a GRF from the Research Grants Council of Hong Kong (Ref.: 11205620). + +# References + +[1] Omri Avrahami, Thomas Hayes, Oran Gafni, Sonal Gupta, Yaniv Taigman, Devi Parikh, Dani Lischinski, Ohad Fried, and Xi Yin. Spatext: Spatio-textual representation for controllable image generation. In CVPR, pages 18370-18380, 2023. 3 +[2] Yogesh Balaji, Seungjun Nah, Xun Huang, Arash Vahdat, Jiaming Song, Qinsheng Zhang, Karsten Kreis, Miika Aittala, Timo Aila, Samuli Laine, et al. ediff-i: Text-to-image diffusion models with an ensemble of expert denoisers. arXiv:2211.01324, 2022. 3.5 +[3] Omer Bar-Tal, Lior Yariv, Yaron Lipman, and Tali Dekel. Multidiffusion: fusing diffusion paths for controlled image generation. In ICML, pages 1737-1752, 2023. 2, 3, 7, 5 +[4] Tim Brooks, Aleksander Holynski, and Alexei A Efros. Instructpix2pix: Learning to follow image editing instructions. In CVPR, pages 18392-18402, 2023. 2 +[5] Hila Chefer, Yuval Alaluf, Yael Vinker, Lior Wolf, and Daniel Cohen-Or. Attend-and-excite: Attention-based semantic guidance for text-to-image diffusion models. ACM TOG, 42(4):1-10, 2023. 3, 1 +[6] Hong Chen, Yipeng Zhang, Xin Wang, Xuguang Duan, Yuwei Zhou, and Wenwu Zhu. Disenbooth: Disentangled parameter-efficient tuning for subject-driven text-to-image generation. arXiv:2305.03374, 3, 2023. 1 +[7] Kai Chen, Enze Xie, Zhe Chen, Yibo Wang, HONG Lanqing, Zhenguo Li, and Dit-Yan Yeung. Geodiffusion: Text-prompted geometric control for object detection data generation. In ICLR, 2024. 3 +[8] Minghao Chen, Iro Laina, and Andrea Vedaldi. Training-free layout control with cross-attention guidance. In WACV, pages 5343-5353, 2024. 3, 2 +[9] Xi Chen, Lianghua Huang, Yu Liu, Yujun Shen, Deli Zhao, and Hengshuang Zhao. Anydoor: Zero-shot object-level image customization. In CVPR, pages 6593-6602, 2024. 3, 1, 6 +[10] Jiaxin Cheng, Xiao Liang, Xingjian Shi, Tong He, Tianjun Xiao, and Mu Li. Layoutdiffuse: Adapting foundational diffusion models for layout-to-image generation. arXiv preprint arXiv:2302.08908, 2023. 3 +[11] Patrick Esser, Robin Rombach, and Bjorn Ommer. Taming transformers for high-resolution image synthesis. In CVPR, pages 12873-12883, 2021. 3 +[12] Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. In ICML, 2024. 3 +[13] Weixi Feng, Xuehai He, Tsu-Jui Fu, Varun Jampani, Arjun Akula, Pradyumna Narayana, Sugato Basu, Xin Eric + +Wang, and William Yang Wang. Training-free structured diffusion guidance for compositional text-to-image synthesis. arXiv:2212.05032, 2022. 3 +[14] Rinon Gal, Yuval Alaluf, Yuval Atzmon, Or Patashnik, Amit H Bermano, Gal Chechik, and Daniel Cohen-Or. An image is worth one word: Personalizing text-to-image generation using textual inversion. arXiv:2208.01618, 2022. 1 +[15] Biao Gong, Siteng Huang, Yutong Feng, Shiwei Zhang, Yuyuan Li, and Yu Liu. Check locate rectify: A training-free layout calibration system for text-to-image generation. In CVPR, pages 6624-6634, 2024. 3 +[16] Shuyang Gu, Dong Chen, Jianmin Bao, Fang Wen, Bo Zhang, Dongdong Chen, Lu Yuan, and Baining Guo. Vector quantized diffusion model for text-to-image synthesis. In CVPR, pages 10696-10706, 2022. 3 +[17] Amir Hertz, Ron Mokady, Jay Tenenbaum, Kfir Aberman, Yael Pritch, and Daniel Cohen-or. Prompt-to-prompt image editing with cross-attention control. In ICLR, 2023. 5 +[18] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In NIPS, 2017. 6 +[19] Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. In NIPS Workshop, 2021. 5 +[20] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In NIPS, 2020. 2, 3 +[21] Jonathan Ho, Chitwan Sahara, William Chan, David J Fleet, Mohammad Norouzi, and Tim Salimans. Cascaded diffusion models for high fidelity image generation. JMLR, 23(47):1-33, 2022. 3 +[22] Hexiang Hu, Kelvin CK Chan, Yu-Chuan Su, Wenhu Chen, Yandong Li, Kihyuk Sohn, Yang Zhao, Xue Ben, Boqing Gong, William Cohen, et al. Instruct-imagen: Image generation with multi-modal instruction. In CVPR, pages 4754-4763, 2024. 3 +[23] Chengyou Jia, Minnan Luo, Zhuohang Dang, Guang Dai, Xiaojun Chang, Mengmeng Wang, and Jingdong Wang. Ssmg: Spatial-semantic map guided diffusion model for free-form layout-to-image generation. In AAAI, pages 2480–2488, 2024. 3 +[24] Yuming Jiang, Nanxuan Zhao, Qing Liu, Krishna Kumar Singh, Shuai Yang, Chen Change Loy, and Ziwei Liu. Groupdiff: Diffusion-based group portrait editing. arXiv:2409.14379, 2024. 1 +[25] Zeyinzi Jiang, Chaojie Mao, Yulin Pan, Zhen Han, and Jingfeng Zhang. Scedit: Efficient and controllable image diffusion generation via skip connection editing. In CVPR, pages 8995–9004, 2024. 3 +[26] Sungnyun Kim, Junsoo Lee, Kibeom Hong, Daesik Kim, and Namhyuk Ahn. Diffblender: Scalable and composable multimodal text-to-image diffusion models. arXiv:2305.15194, 2023. 3 +[27] Yunji Kim, Jiyoung Lee, Jin-Hwa Kim, Jung-Woo Ha, and Jun-Yan Zhu. Dense text-to-image generation with attention modulation. In ICCV, pages 7701-7711, 2023. 1, 2, 3, 4, 7, 5 + +[28] Leheng Li, Weichao Qiu, Xu Yan, Jing He, Kaiqiang Zhou, Yingjie Cai, Qing Lian, Bingbing Liu, and Ying-Cong Chen. Omnibooth: Learning latent control for image synthesis with multi-modal instruction. arXiv:2410.04932, 2024. 1 +[29] Tianle Li, Max Ku, Cong Wei, and Wenhu Chen. Dreamedit: Subject-driven image editing. arXiv:2306.12624, 2023. 3, 1 +[30] Yuheng Li, Haotian Liu, Qingyang Wu, Fangzhou Mu, Jianwei Yang, Jianfeng Gao, Chunyuan Li, and Yong Jae Lee. Gligen: Open-set grounded text-to-image generation. In CVPR, pages 22511-22521, 2023. 1, 2, 3, 8, 5, 6, 9 +[31] Yaqi Li, Han Fang, Zerun Feng, Kaijing Ma, Chao Ban, Xianghao Zang, LanXiang Zhou, Zhongjiang He, Jingyan Chen, Jiani Hu, et al. Goal: Grounded text-to-image synthesis with joint layout alignment tuning. In ACM MM, pages 7055-7064, 2024. 3 +[32] Chang Liu, Xiangtai Li, and Henghui Ding. Referring image editing: Object-level image editing via referring expressions. In CVPR, pages 13128-13138, 2024. 3, 1 +[33] Nan Liu, Shuang Li, Yilun Du, Antonio Torralba, and Joshua B Tenenbaum. Compositional visual generation with composable diffusion models. In ECCV, pages 423-439, 2022. 3 +[34] Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. arXiv:2303.05499, 2023.6.1 +[35] Xiaoyu Liu, Yuxiang Wei, Ming Liu, Xianhui Lin, Peiran Ren, Xuansong Xie, and Wangmeng Zuo. Smartcontrol: Enhancing controlnet for handling rough visual conditions. arXiv:2404.06451, 2024. 3, 7 +[36] Yuhao Liu, Zhanghan Ke, Fang Liu, Nanxuan Zhao, and Rynson WH Lau. Diff-plugin: Revitalizing details for diffusion-based low-level tasks. In CVPR, pages 4197-4208, 2024. 3 +[37] Grace Luo, Trevor Darrell, Oliver Wang, Dan B Goldman, and Aleksander Holynski. Readout guidance: Learning control from diffusion features. In CVPR, pages 8217-8227, 2024. 3 +[38] Kangfu Mei, Mauricio Delbracio, Hossein Talebi, Zhengzhong Tu, Vishal M Patel, and Peyman Milanfar. Codi: Conditional diffusion distillation for higher-fidelity and faster image generation. In CVPR, pages 9048-9058, 2024. 3 +[39] Sicheng Mo, Fangzhou Mu, Kuan Heng Lin, Yanli Liu, Bochen Guan, Yin Li, and Bolei Zhou. Freecontrol: Training-free spatial control of any text-to-image diffusion model with any condition. In CVPR, pages 7465-7475, 2024. 3, 4, 6, 7, 5 +[40] Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In AAAI, pages 4296-4304, 2024. 3 +[41] Nithin Gopalakrishnan Nair, Jeya Maria Jose Valanarasu, and Vishal M Patel. Maxfusion: Plug&play multi-modal generation in text-to-image diffusion models. In ECCV, pages 93-110, 2025. 3 + +[42] Martin E Newell, RG Newell, and Tom L Sancha. A solution to the hidden surface problem. In Seminal graphics: pioneering efforts that shaped the field, pages 27-32. 1998. 4 +[43] Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. Glide: Towards photorealistic image generation and editing with text-guided diffusion models. arXiv:2112.10741, 2021. 3 +[44] NovelAI. Novelai: Ai generated stories and art. 2021. 2 +[45] Lingzhi Pan, Tong Zhang, Bingyuan Chen, Qi Zhou, Wei Ke, Sabine Süssstrunk, and Mathieu Salzmann. Coherent and multi-modality image inpainting via latent space optimization. arXiv:2407.08019, 2024. 1 +[46] Yulin Pan, Chaojie Mao, Zeyinzi Jiang, Zhen Han, and Jingfeng Zhang. Locate, assign, refine: Taming customized image inpainting with text-subject guidance. arXiv:2403.19534, 2024. 1 +[47] Quynh Phung, Songwei Ge, and Jia-Bin Huang. Grounded text-to-image synthesis with attention refocusing. In CVPR, pages 7932-7942, 2024. 2, 3, 4, 6, 7, 5 +[48] Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv:2307.01952, 2023. 3, 5 +[49] Zipeng Qi, Guoxi Huang, Chenyang Liu, and Fei Ye. Layered rendering diffusion model for zero-shot guided image synthesis. In ECCV, 2024. 3, 1 +[50] Can Qin, Shu Zhang, Ning Yu, Yihao Feng, Xinyi Yang, Yingbo Zhou, Huan Wang, Juan Carlos Niebles, Caiming Xiong, Silvio Savarese, et al. Uniconcontrol: A unified diffusion model for controllable visual generation in the wild. In NIPS, 2023. 3 +[51] Leigang Qu, Shengqiong Wu, Hao Fei, Liqiang Nie, and TatSeng Chua. Layoutllm-t2i: Eliciting layout guidance from llm for text-to-image generation. In ACM MM, pages 643-654, 2023. 3 +[52] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In ICML, pages 8748-8763, 2021. 3, 4 +[53] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. JMLR, 21(140):1-67, 2020. 3 +[54] Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. arXiv:2204.06125, 2022. 2, 3 +[55] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In CVPR, pages 10684-10695, 2022. 1, 2, 3, 4, 5 +[56] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine + +tuning text-to-image diffusion models for subject-driven generation. In CVPR, pages 22500-22510, 2023. 8, 1, 6 +[57] Chitwan Sahara, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding. In NIPS, pages 36479-36494, 2022. 2, 3 +[58] Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Wortsman, et al. Laion-5b: An open large-scale dataset for training next generation image-text models. In NIPS, pages 25278–25294, 2022. 3, 1 +[59] Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. In ICLR, 2021. 2, 3, 8, 1, 6 +[60] Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In ICLR, 2021. 2, 3 +[61] Yizhi Song, Zhifei Zhang, Zhe Lin, Scott Cohen, Brian Price, Jianming Zhang, Soo Ye Kim, and Daniel Aliaga. Object-stitch: Object compositing with diffusion model. In CVPR, pages 18310-18319, 2023. 3, 1 +[62] Yizhi Song, Zhifei Zhang, Zhe Lin, Scott Cohen, Brian Price, Jianming Zhang, Soo Ye Kim, He Zhang, Wei Xiong, and Daniel Aliaga. Imprint: Generative object compositing by learning identity-preserving representation. In CVPR, pages 8048-8058, 2024. 3, 1, 6 +[63] Yanan Sun, Yanchen Liu, Yinhao Tang, Wenjie Pei, and Kai Chen. Anycontrol: Create your artwork with versatile control on text-to-image generation, 2024. 3, 6, 7 +[64] Zhenhong Sun, Junyan Wang, Zhiyu Tan, Daoyi Dong, Hailan Ma, Hao Li, and Dong Gong. Eggen: Image generation with multi-entity prior learning through entity guidance. In ACM MM, pages 6637-6645, 2024. 3 +[65] Xudong Wang, Trevor Darrell, Sai Saketh Rambhatla, Rohit Girdhar, and Ishan Misra. Instancediffusion: Instance-level control for image generation. In CVPR, pages 6232-6242, 2024. 3, 6, 7, 8, 5, 9 +[66] Yuxiang Wei, Yabo Zhang, Zhilong Ji, Jinfeng Bai, Lei Zhang, and Wangmeng Zuo. Elite: Encoding visual concepts into textual embeddings for customized text-to-image generation. In ICCV, pages 15943-15953, 2023. 1 +[67] Yinwei Wu, Xianpan Zhou, Bing Ma, Xuefeng Su, Kai Ma, and Xinchao Wang. Ifadapter: Instance feature control for grounded text-to-image generation. arXiv:2409.08240, 2024.3 +[68] Jiayu Xiao, Henglei Lv, Liang Li, Shuhui Wang, and Qingming Huang. R&b: Region and boundary aware zero-shot grounded text-to-image generation. In ICLR, 2023. 2, 3, 6, 7, 5 +[69] Shitao Xiao, Yueze Wang, Junjie Zhou, Huaying Yuan, Xingrun Xing, Ruiran Yan, Shuting Wang, Tiejun Huang, and Zheng Liu. Omnigen: Unified image generation. arXiv:2409.11340, 2024. 3 + +[70] Jinheng Xie, Yuexiang Li, Yawen Huang, Haozhe Liu, Wentian Zhang, Yefeng Zheng, and Mike Zheng Shou. Boxdiff: Text-to-image synthesis with training-free box-constrained diffusion. In ICCV, pages 7452-7461, 2023. 2, 3, 6, 7, 5 +[71] Xingqian Xu, Jiayi Guo, Zhangyang Wang, Gao Huang, Irfan Essa, and Humphrey Shi. Prompt-free diffusion: Taking" text" out of text-to-image diffusion models. In CVPR, pages 8682-8692, 2024. 3 +[72] Binxin Yang, Shuyang Gu, Bo Zhang, Ting Zhang, Xuejin Chen, Xiaoyan Sun, Dong Chen, and Fang Wen. Paint by example: Exemplar-based image editing with diffusion models. In CVPR, pages 18381-18391, 2023. 3, 1, 6 +[73] Ling Yang, Zhaochen Yu, Chenlin Meng, Minkai Xu, Stefano Ermon, and CUI Bin. Mastering text-to-image diffusion: Recaptioning, planning, and generating with multimodal llms. In ICML, 2024. 3, 6, 7, 5 +[74] Ling Yang, Zhilong Zhang, Zhaochen Yu, Jingwei Liu, Minkai Xu, Stefano Ermon, and Bin Cui. Cross-modal contextualized diffusion models for text-guided visual generation and editing. arXiv:2402.16627, 2024. 3 +[75] Yang Yang, Wen Wang, Liang Peng, Chaotian Song, Yao Chen, Hengjia Li, Xiaolong Yang, Qinglin Lu, Deng Cai, Boxi Wu, et al. Lora-composer: Leveraging low-rank adaptation for multi-concept customization in training-free diffusion models. arXiv:2403.11627, 2024. 1 +[76] Zhengyuan Yang, Jianfeng Wang, Zhe Gan, Linjie Li, Kevin Lin, Chenfei Wu, Nan Duan, Zicheng Liu, Ce Liu, Michael Zeng, et al. Reco: Region-controlled text-to-image generation. In CVPR, pages 14246-14255, 2023. 3 +[77] Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv:2308.06721, 2023. 1 +[78] Ziyang Yuan, Mingdeng Cao, Xintao Wang, Zhonggang Qi, Chun Yuan, and Ying Shan. Customnet: Zero-shot object customization with variable-viewpoints in text-to-image diffusion models. arXiv:2310.19784, 2023. 1 +[79] Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In ICCV, pages 3836-3847, 2023. 3, 6, 7, 8, 5, 9 +[80] Lirui Zhao, Tianshuo Yang, Wenqi Shao, Yuxin Zhang, Yu Qiao, Ping Luo, Kaipeng Zhang, and Rongrong Ji. Diffree: Text-guided shape free object inpainting with diffusion model. arXiv:2407.16982, 2024. 3, 1 +[81] Shihao Zhao, Dongdong Chen, Yen-Chun Chen, Jianmin Bao, Shaozhe Hao, Lu Yuan, and Kwan-Yee K Wong. Uni-controlnet: All-in-one control to text-to-image diffusion models. In NIPS, 2023. 3 +[82] Dewei Zhou, You Li, Fan Ma, Zongxin Yang, and Yi Yang. Mige $^{+ + }$ : Advanced multi-instance generation controller for image synthesis. arXiv:2407.02329, 2024. 3 +[83] Dewei Zhou, You Li, Fan Ma, Xiaoting Zhang, and Yi Yang. Mige: Multi-instance generation controller for text-to-image synthesis. In CVPR, pages 6818-6828, 2024. 1, 2, 3, 6, 7, 8, 5, 9 \ No newline at end of file diff --git a/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/images.zip b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/images.zip new file mode 100644 index 0000000000000000000000000000000000000000..a8c71151866e6d57d5f48f7ab7af87a948b27c7f --- /dev/null +++ b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/images.zip @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb0d976c972f2796dac6214a643fe2176446d694a1c63f6eb8848cdde834337f +size 692842 diff --git a/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/layout.json b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/layout.json new file mode 100644 index 0000000000000000000000000000000000000000..066a089eef9d82c39700a0559ebaec3da683a5b6 --- /dev/null +++ b/2025/VODiff_ Controlling Object Visibility Order in Text-to-Image Generation/layout.json @@ -0,0 +1,14080 @@ +{ + "pdf_info": [ + { + "para_blocks": [ + { + "bbox": [ + 78, + 103, + 532, + 121 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 78, + 103, + 532, + 121 + ], + "spans": [ + { + "bbox": [ + 78, + 103, + 532, + 121 + ], + "type": "text", + "content": "VODiff: Controlling Object Visibility Order in Text-to-Image Generation" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "spans": [ + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "text", + "content": "Dong Liang" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "inline_equation", + "content": "^{1,2}" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "text", + "content": " Jinyuan Jia" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "inline_equation", + "content": "^{1,3,\\dagger}" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "text", + "content": " Yuhao Liu" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "inline_equation", + "content": "^{2,\\dagger}" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "text", + "content": " Zhanghan Ke" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "inline_equation", + "content": "^{2}" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "text", + "content": " Hongbo Fu" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "inline_equation", + "content": "^{4}" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "text", + "content": " Rynson W.H. Lau" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "inline_equation", + "content": "^{2,\\dagger}" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "inline_equation", + "content": "^{1}" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "text", + "content": "Tongji University " + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "inline_equation", + "content": "^{2}" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "text", + "content": "City University of Hong Kong " + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "inline_equation", + "content": "^{3}" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "text", + "content": "HKUST(GZ) " + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "inline_equation", + "content": "^{4}" + }, + { + "bbox": [ + 61, + 147, + 556, + 180 + ], + "type": "text", + "content": "HKUST" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 91, + 182, + 522, + 208 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 91, + 182, + 522, + 208 + ], + "spans": [ + { + "bbox": [ + 91, + 182, + 522, + 208 + ], + "type": "text", + "content": "sse_liangdong@tongji.edu.cn, jinyuanjia@hkust-gz.edu.cn, yuhaoliu7456@gmail.com, zhanghake2-c@my.cityu.edu.hk, hongbofu@ust.hk, Rynson.Lau@cityu.edu.hk" + } + ] + } + ], + "index": 4 + }, + { + "type": "image", + "bbox": [ + 59, + 237, + 134, + 312 + ], + "blocks": [ + { + "bbox": [ + 66, + 224, + 126, + 235 + ], + "lines": [ + { + "bbox": [ + 66, + 224, + 126, + 235 + ], + "spans": [ + { + "bbox": [ + 66, + 224, + 126, + 235 + ], + "type": "text", + "content": "(a) Layout+Order" + } + ] + } + ], + "index": 5, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 59, + 237, + 134, + 312 + ], + "lines": [ + { + "bbox": [ + 59, + 237, + 134, + 312 + ], + "spans": [ + { + "bbox": [ + 59, + 237, + 134, + 312 + ], + "type": "image", + "image_path": "605eb3a26f0b7f7ec9bc5363a4f2b3c1fb8f14102a074ea03cac17478070f9e8.jpg" + } + ] + } + ], + "index": 6, + "angle": 0, + "type": "image_body" + } + ], + "index": 6 + }, + { + "type": "image", + "bbox": [ + 145, + 238, + 219, + 312 + ], + "blocks": [ + { + "bbox": [ + 158, + 224, + 209, + 235 + ], + "lines": [ + { + "bbox": [ + 158, + 224, + 209, + 235 + ], + "spans": [ + { + "bbox": [ + 158, + 224, + 209, + 235 + ], + "type": "text", + "content": "(b) SD 1.5 [55]" + } + ] + } + ], + "index": 7, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 145, + 238, + 219, + 312 + ], + "lines": [ + { + "bbox": [ + 145, + 238, + 219, + 312 + ], + "spans": [ + { + "bbox": [ + 145, + 238, + 219, + 312 + ], + "type": "image", + "image_path": "396ae77a63730ad6b504e4a40fbeb46cb8acba828a8beb07b5bc76d08f7ca37b.jpg" + } + ] + } + ], + "index": 8, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 171, + 314, + 447, + 326 + ], + "lines": [ + { + "bbox": [ + 171, + 314, + 447, + 326 + ], + "spans": [ + { + "bbox": [ + 171, + 314, + 447, + 326 + ], + "type": "text", + "content": "\"A picture of a table, with a mug and a cup on it, the mug is on a dish.\"" + } + ] + } + ], + "index": 17, + "angle": 0, + "type": "image_footnote" + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 223, + 238, + 296, + 312 + ], + "blocks": [ + { + "bbox": [ + 232, + 224, + 290, + 235 + ], + "lines": [ + { + "bbox": [ + 232, + 224, + 290, + 235 + ], + "spans": [ + { + "bbox": [ + 232, + 224, + 290, + 235 + ], + "type": "text", + "content": "(c) GLIGEN [30]" + } + ] + } + ], + "index": 9, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 223, + 238, + 296, + 312 + ], + "lines": [ + { + "bbox": [ + 223, + 238, + 296, + 312 + ], + "spans": [ + { + "bbox": [ + 223, + 238, + 296, + 312 + ], + "type": "image", + "image_path": "90d8390c805022ce98b0af3bd8a069e40e530d985c5c3495f8e28953b43c651b.jpg" + } + ] + } + ], + "index": 10, + "angle": 0, + "type": "image_body" + } + ], + "index": 10 + }, + { + "type": "image", + "bbox": [ + 299, + 238, + 376, + 312 + ], + "blocks": [ + { + "bbox": [ + 299, + 224, + 378, + 235 + ], + "lines": [ + { + "bbox": [ + 299, + 224, + 378, + 235 + ], + "spans": [ + { + "bbox": [ + 299, + 224, + 378, + 235 + ], + "type": "text", + "content": "(d) DenseDiffusion [27]" + } + ] + } + ], + "index": 11, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 299, + 238, + 376, + 312 + ], + "lines": [ + { + "bbox": [ + 299, + 238, + 376, + 312 + ], + "spans": [ + { + "bbox": [ + 299, + 238, + 376, + 312 + ], + "type": "image", + "image_path": "9156d041384b67c029095cc08a48ac276a6775d1663f900496069dd8f81bacb8.jpg" + } + ] + } + ], + "index": 12, + "angle": 0, + "type": "image_body" + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 378, + 238, + 455, + 312 + ], + "blocks": [ + { + "bbox": [ + 394, + 224, + 443, + 235 + ], + "lines": [ + { + "bbox": [ + 394, + 224, + 443, + 235 + ], + "spans": [ + { + "bbox": [ + 394, + 224, + 443, + 235 + ], + "type": "text", + "content": "(e) MIGC [83]" + } + ] + } + ], + "index": 13, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 378, + 238, + 455, + 312 + ], + "lines": [ + { + "bbox": [ + 378, + 238, + 455, + 312 + ], + "spans": [ + { + "bbox": [ + 378, + 238, + 455, + 312 + ], + "type": "image", + "image_path": "3ce347d46bf265f8084f29a7a91ae70c3642042c6d1324a0846eac8aae94d125.jpg" + } + ] + } + ], + "index": 14, + "angle": 0, + "type": "image_body" + } + ], + "index": 14 + }, + { + "type": "image", + "bbox": [ + 467, + 238, + 542, + 312 + ], + "blocks": [ + { + "bbox": [ + 475, + 224, + 534, + 235 + ], + "lines": [ + { + "bbox": [ + 475, + 224, + 534, + 235 + ], + "spans": [ + { + "bbox": [ + 475, + 224, + 534, + 235 + ], + "type": "text", + "content": "(f) VODiff (Ours)" + } + ] + } + ], + "index": 15, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 467, + 238, + 542, + 312 + ], + "lines": [ + { + "bbox": [ + 467, + 238, + 542, + 312 + ], + "spans": [ + { + "bbox": [ + 467, + 238, + 542, + 312 + ], + "type": "image", + "image_path": "2e1e30381a2348527e12af9930f5f690daeb7d743f2ff622f2309da400d2597f.jpg" + } + ] + } + ], + "index": 16, + "angle": 0, + "type": "image_body" + } + ], + "index": 16 + }, + { + "type": "image", + "bbox": [ + 59, + 331, + 134, + 407 + ], + "blocks": [ + { + "bbox": [ + 59, + 331, + 134, + 407 + ], + "lines": [ + { + "bbox": [ + 59, + 331, + 134, + 407 + ], + "spans": [ + { + "bbox": [ + 59, + 331, + 134, + 407 + ], + "type": "image", + "image_path": "7d2c4545c2adaf30a0ee372b5f280c38b4cfbe7e01ad5f5c7824507005d7b049.jpg" + } + ] + } + ], + "index": 18, + "angle": 0, + "type": "image_body" + } + ], + "index": 18 + }, + { + "type": "image", + "bbox": [ + 145, + 332, + 219, + 407 + ], + "blocks": [ + { + "bbox": [ + 145, + 332, + 219, + 407 + ], + "lines": [ + { + "bbox": [ + 145, + 332, + 219, + 407 + ], + "spans": [ + { + "bbox": [ + 145, + 332, + 219, + 407 + ], + "type": "image", + "image_path": "d002f4801e9720ae91a0bb523b77c2a4d91a051492d21f0091802344fc7b52a5.jpg" + } + ] + } + ], + "index": 19, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 191, + 409, + 425, + 421 + ], + "lines": [ + { + "bbox": [ + 191, + 409, + 425, + 421 + ], + "spans": [ + { + "bbox": [ + 191, + 409, + 425, + 421 + ], + "type": "text", + "content": "\"A suitcase with a banana in front of it and a dog behind it.\"" + } + ] + } + ], + "index": 24, + "angle": 0, + "type": "image_footnote" + } + ], + "index": 19 + }, + { + "type": "image", + "bbox": [ + 223, + 332, + 296, + 406 + ], + "blocks": [ + { + "bbox": [ + 223, + 332, + 296, + 406 + ], + "lines": [ + { + "bbox": [ + 223, + 332, + 296, + 406 + ], + "spans": [ + { + "bbox": [ + 223, + 332, + 296, + 406 + ], + "type": "image", + "image_path": "74686a116326c0a5784c05a7b5fd42838d1ce6c0c438688e1adcf37242e2f0e1.jpg" + } + ] + } + ], + "index": 20, + "angle": 0, + "type": "image_body" + } + ], + "index": 20 + }, + { + "type": "image", + "bbox": [ + 299, + 332, + 376, + 406 + ], + "blocks": [ + { + "bbox": [ + 299, + 332, + 376, + 406 + ], + "lines": [ + { + "bbox": [ + 299, + 332, + 376, + 406 + ], + "spans": [ + { + "bbox": [ + 299, + 332, + 376, + 406 + ], + "type": "image", + "image_path": "2c8395177c695dd085d36e42b5856418d2e985a9761daf6162d4040bdd94d995.jpg" + } + ] + } + ], + "index": 21, + "angle": 0, + "type": "image_body" + } + ], + "index": 21 + }, + { + "type": "image", + "bbox": [ + 378, + 332, + 455, + 407 + ], + "blocks": [ + { + "bbox": [ + 378, + 332, + 455, + 407 + ], + "lines": [ + { + "bbox": [ + 378, + 332, + 455, + 407 + ], + "spans": [ + { + "bbox": [ + 378, + 332, + 455, + 407 + ], + "type": "image", + "image_path": "ad07d422a8159eb26d5d54266d112d32414f2bd332d94e9767163a9e3dc248b6.jpg" + } + ] + } + ], + "index": 22, + "angle": 0, + "type": "image_body" + } + ], + "index": 22 + }, + { + "type": "image", + "bbox": [ + 468, + 333, + 542, + 407 + ], + "blocks": [ + { + "bbox": [ + 468, + 333, + 542, + 407 + ], + "lines": [ + { + "bbox": [ + 468, + 333, + 542, + 407 + ], + "spans": [ + { + "bbox": [ + 468, + 333, + 542, + 407 + ], + "type": "image", + "image_path": "0e8847e8ce613ebf337399df08f281f5452a53b69c60edaf0d07a85bf2af47d3.jpg" + } + ] + } + ], + "index": 23, + "angle": 0, + "type": "image_body" + } + ], + "index": 23 + }, + { + "type": "image", + "bbox": [ + 59, + 428, + 134, + 503 + ], + "blocks": [ + { + "bbox": [ + 59, + 428, + 134, + 503 + ], + "lines": [ + { + "bbox": [ + 59, + 428, + 134, + 503 + ], + "spans": [ + { + "bbox": [ + 59, + 428, + 134, + 503 + ], + "type": "image", + "image_path": "e0cd9df4466b5e25d275006dd0334a0aaafe8427b37a82dd13563634a7e6fca4.jpg" + } + ] + } + ], + "index": 25, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 130, + 506, + 492, + 518 + ], + "lines": [ + { + "bbox": [ + 130, + 506, + 492, + 518 + ], + "spans": [ + { + "bbox": [ + 130, + 506, + 492, + 518 + ], + "type": "text", + "content": "\"A picture of a living room with a table and a plant in front of a sofa, and a dog on the sofa.\"" + } + ] + } + ], + "index": 31, + "angle": 0, + "type": "image_footnote" + }, + { + "bbox": [ + 55, + 522, + 555, + 567 + ], + "lines": [ + { + "bbox": [ + 55, + 522, + 555, + 567 + ], + "spans": [ + { + "bbox": [ + 55, + 522, + 555, + 567 + ], + "type": "text", + "content": "Figure 1. Existing T2I methods that rely on text prompts (b) and those that combine text and layout conditioning (c-e) often struggle to produce images with accurate occlusion relationships. In this work, we propose a new framework called VODiff, which enhances control over object arrangement by introducing their visibility orders (indicated by numbers above their bounding box in the layout map in (a)) as auxiliary constraints. VODiff enables the generation of images with correct spatial arrangements and occlusion relationships." + } + ] + } + ], + "index": 32, + "angle": 0, + "type": "image_caption" + } + ], + "index": 25 + }, + { + "type": "image", + "bbox": [ + 145, + 429, + 219, + 503 + ], + "blocks": [ + { + "bbox": [ + 145, + 429, + 219, + 503 + ], + "lines": [ + { + "bbox": [ + 145, + 429, + 219, + 503 + ], + "spans": [ + { + "bbox": [ + 145, + 429, + 219, + 503 + ], + "type": "image", + "image_path": "cd961ee444dbf23977023005ab0f2dd6e8abeba45d987059b2b9ffabc9e27a6f.jpg" + } + ] + } + ], + "index": 26, + "angle": 0, + "type": "image_body" + } + ], + "index": 26 + }, + { + "type": "image", + "bbox": [ + 222, + 429, + 295, + 503 + ], + "blocks": [ + { + "bbox": [ + 222, + 429, + 295, + 503 + ], + "lines": [ + { + "bbox": [ + 222, + 429, + 295, + 503 + ], + "spans": [ + { + "bbox": [ + 222, + 429, + 295, + 503 + ], + "type": "image", + "image_path": "ba94a955f8739f650d20019a5b7ee45b70244a8f863f81f790cb4325b2158ee3.jpg" + } + ] + } + ], + "index": 27, + "angle": 0, + "type": "image_body" + } + ], + "index": 27 + }, + { + "type": "image", + "bbox": [ + 299, + 429, + 376, + 503 + ], + "blocks": [ + { + "bbox": [ + 299, + 429, + 376, + 503 + ], + "lines": [ + { + "bbox": [ + 299, + 429, + 376, + 503 + ], + "spans": [ + { + "bbox": [ + 299, + 429, + 376, + 503 + ], + "type": "image", + "image_path": "7c21e606ccfd361c39ecff52685c2c865629e60a794729a3d1840f630f373b41.jpg" + } + ] + } + ], + "index": 28, + "angle": 0, + "type": "image_body" + } + ], + "index": 28 + }, + { + "type": "image", + "bbox": [ + 378, + 429, + 456, + 503 + ], + "blocks": [ + { + "bbox": [ + 378, + 429, + 456, + 503 + ], + "lines": [ + { + "bbox": [ + 378, + 429, + 456, + 503 + ], + "spans": [ + { + "bbox": [ + 378, + 429, + 456, + 503 + ], + "type": "image", + "image_path": "3d4ca8009b08819f411e09f0503f73962549d923b4cc04d31f104db0bc7869c4.jpg" + } + ] + } + ], + "index": 29, + "angle": 0, + "type": "image_body" + } + ], + "index": 29 + }, + { + "type": "image", + "bbox": [ + 467, + 429, + 542, + 504 + ], + "blocks": [ + { + "bbox": [ + 467, + 429, + 542, + 504 + ], + "lines": [ + { + "bbox": [ + 467, + 429, + 542, + 504 + ], + "spans": [ + { + "bbox": [ + 467, + 429, + 542, + 504 + ], + "type": "image", + "image_path": "b363900ca272bb5e12e8a111fb2876bbf094d57e1f4087ab4f293a320cf2acb2.jpg" + } + ] + } + ], + "index": 30, + "angle": 0, + "type": "image_body" + } + ], + "index": 30 + }, + { + "bbox": [ + 151, + 581, + 200, + 594 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 151, + 581, + 200, + 594 + ], + "spans": [ + { + "bbox": [ + 151, + 581, + 200, + 594 + ], + "type": "text", + "content": "Abstract" + } + ] + } + ], + "index": 33 + }, + { + "bbox": [ + 55, + 608, + 295, + 692 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 608, + 295, + 692 + ], + "spans": [ + { + "bbox": [ + 55, + 608, + 295, + 692 + ], + "type": "text", + "content": "Recent advancements in diffusion models have significantly enhanced the performance of text-to-image models in image synthesis. To enable control over the spatial locations of the generated objects, diffusion-based methods typically utilize object layout as an auxiliary input. However, we observe that this approach treats all objects as being on the same layer and neglect their visibility order, lead" + } + ] + } + ], + "index": 34 + }, + { + "bbox": [ + 313, + 582, + 555, + 702 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 582, + 555, + 702 + ], + "spans": [ + { + "bbox": [ + 313, + 582, + 555, + 702 + ], + "type": "text", + "content": "ing to the synthesis of overlapping objects with incorrect occlusions. To address this limitation, we introduce in this paper a new training-free framework that considers object visibility order explicitly and allows users to place overlapping objects in a stack of layers. Our framework consists of two visibility-based designs. First, we propose a novel Sequential Denoising Process (SDP) to divide the whole image generation into multiple stages for different objects, each stage primarily focuses on an object. Second, we propose a novel Visibility-Order-Aware (VOA) Loss to trans" + } + ] + } + ], + "index": 35 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "spans": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "type": "text", + "content": "CVF" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 145, + 0, + 496, + 37 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 145, + 0, + 496, + 37 + ], + "spans": [ + { + "bbox": [ + 145, + 0, + 496, + 37 + ], + "type": "text", + "content": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 70, + 702, + 167, + 713 + ], + "type": "page_footnote", + "angle": 0, + "lines": [ + { + "bbox": [ + 70, + 702, + 167, + 713 + ], + "spans": [ + { + "bbox": [ + 70, + 702, + 167, + 713 + ], + "type": "text", + "content": "†Joint corresponding authors." + } + ] + } + ], + "index": 36 + }, + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "text", + "content": "18379" + } + ] + } + ], + "index": 37 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 0 + }, + { + "para_blocks": [ + { + "bbox": [ + 54, + 72, + 294, + 204 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 72, + 294, + 204 + ], + "spans": [ + { + "bbox": [ + 54, + 72, + 294, + 204 + ], + "type": "text", + "content": "form the layout and occlusion constraints into an attention map optimization process to improve the accuracy of synthesizing object occlusions in complex scenes. By merging these two novel components, our framework, dubbed VODiff, enables the generation of photorealistic images that satisfy user-specified spatial constraints and object occlusion relationships. In addition, we introduce VOBench, a diverse benchmark dataset containing 200 curated samples, each with a reference image, text prompts, object visibility orders and layout maps. We conduct extensive evaluations on this dataset to demonstrate the superiority of our approach." + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 227, + 136, + 239 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 227, + 136, + 239 + ], + "spans": [ + { + "bbox": [ + 55, + 227, + 136, + 239 + ], + "type": "text", + "content": "1. Introduction" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 55, + 247, + 295, + 319 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 247, + 295, + 319 + ], + "spans": [ + { + "bbox": [ + 55, + 247, + 295, + 319 + ], + "type": "text", + "content": "With the advancements of diffusion models [20, 59], text-to-image (T2I) models [54, 55, 57] have revolutionized the field of image generation. These large-scale T2I models, e.g., Stable Diffusion [55], can generate diverse, high-quality images conforming to given prompts, facilitating more efficient image editing [4] and artistic creation [44]." + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 319, + 295, + 450 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 319, + 295, + 450 + ], + "spans": [ + { + "bbox": [ + 55, + 319, + 295, + 450 + ], + "type": "text", + "content": "Despite the success, existing T2I models often struggle to produce satisfactory results in complex scenes involving multiple objects with occlusions. On the one hand, text descriptions alone lack spatial controllability, causing text-based methods to likely generate incorrect object positions and relationships, as shown in Fig. 1(b). On the other hand, although some condition-based methods [27, 30, 83] introduce layout conditions (e.g., bbox) into pre-trained T2I models to improve positional accuracy, they do not consider the occlusion relationship of objects and may produce unsatisfactory results, as shown in Fig. 1(c)-(e)." + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 450, + 295, + 653 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 450, + 295, + 653 + ], + "spans": [ + { + "bbox": [ + 55, + 450, + 295, + 653 + ], + "type": "text", + "content": "In this work, we observe that when generating a scene with multiple objects, existing methods typically employ a one-layer-for-all strategy to synthesize all objects simultaneously on a single canvas. Such a design requires the model to associate the objects with the input text description and accurately obtain their spatial/occlusion relationships from it, if available. This is inherently challenging and often results in mixed or even incorrect object arrangements. Even when layout constraints [3, 27, 47, 68, 70] are provided, the model may still struggle with synthesizing the correct visibility order of overlapping objects, resulting in inaccurate occlusions. To address this problem, we propose a new training-free Visibility Order Diffusion (VODiff) framework to explicitly model the visibility order of objects by assigning each user-specified object to a separate layer and executing a novel multi-layer denoising process to produce the output image." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 654, + 295, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 654, + 295, + 713 + ], + "spans": [ + { + "bbox": [ + 55, + 654, + 295, + 713 + ], + "type": "text", + "content": "We first propose a novel Sequential Denoising Process (SDP) to render the input objects one by one. SDP stratifies the objects into distinct layers, from bottom to top, according to the visibility order. For example, in the second example of Fig. 1, the layers are ordered with the \"dog\" at the" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 313, + 72, + 553, + 263 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 72, + 553, + 263 + ], + "spans": [ + { + "bbox": [ + 313, + 72, + 553, + 263 + ], + "type": "text", + "content": "bottom, the \"suitcase\" in the middle, and the \"banana\" at the top. Second, SDP divides the complete denoising process into several sub-stages, with each sub-stage primarily focusing on generating one object. Concurrently, during the denoising process, SDP assigns dynamically varying visual guidance to different objects/layers at different stages using a newly proposed smooth guidance mechanism. To prevent the misalignment of inter-object relations caused by deviations in visual guidance during denoising, we further propose a novel Visibility-Order-Aware (VOA) loss to transform layout and occlusion constraints into attention map optimization. When synthesizing an object, we first divide the corresponding layer into three regions, overlapping, visible, and background regions, based on its visibility w.r.t. those of the other objects, and then apply different constraints on these regions." + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 313, + 265, + 553, + 372 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 265, + 553, + 372 + ], + "spans": [ + { + "bbox": [ + 313, + 265, + 553, + 372 + ], + "type": "text", + "content": "To address the lack of a benchmark in visibility-order-aware T2I generation, we also curate a new benchmark called VOBench, which comprises tailored visibility orders, layouts, referenced images, and prompts. To validate the effectiveness of our VODiff, we conduct extensive evaluations on VOBench. Results show that VODiff generates prompt-consistent objects within the user-specified layout regions and precisely delineates the occlusion relationship, as shown in Fig. 1(f)." + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 326, + 373, + 509, + 384 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 326, + 373, + 509, + 384 + ], + "spans": [ + { + "bbox": [ + 326, + 373, + 509, + 384 + ], + "type": "text", + "content": "Our key contributions can be summarized as:" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 326, + 386, + 553, + 601 + ], + "type": "list", + "angle": 0, + "index": 12, + "blocks": [ + { + "bbox": [ + 326, + 386, + 553, + 445 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 326, + 386, + 553, + 445 + ], + "spans": [ + { + "bbox": [ + 326, + 386, + 553, + 445 + ], + "type": "text", + "content": "- To the best of our knowledge, we are the first to consider visibility order when generating images with T2I models. We present a new training-free framework named VODiff, which explicitly models object visibility order to control object occlusion relationships." + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 326, + 446, + 553, + 529 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 326, + 446, + 553, + 529 + ], + "spans": [ + { + "bbox": [ + 326, + 446, + 553, + 529 + ], + "type": "text", + "content": "- We propose two novel designs to support the explicit modeling of object visibility order: a Sequential Denoising Process (SDP) to generate different objects sequentially at various sub-stages of the denoising process, and a Visibility-Order-Aware (VOA) Loss to incorporate occlusion relationships into the attention optimization process." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 326, + 529, + 553, + 601 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 326, + 529, + 553, + 601 + ], + "spans": [ + { + "bbox": [ + 326, + 529, + 553, + 601 + ], + "type": "text", + "content": "- We curate a visibility-order-aware benchmark called VOBench, and conduct extensive evaluations on it. Results show that compared to existing SOTA methods, VODiff can synthesize more photorealistic images, which satisfy user-specified spatial constraints as well as object visibility order." + } + ] + } + ], + "index": 11 + } + ], + "sub_type": "text" + }, + { + "bbox": [ + 313, + 614, + 400, + 627 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 614, + 400, + 627 + ], + "spans": [ + { + "bbox": [ + 313, + 614, + 400, + 627 + ], + "type": "text", + "content": "2. Related Work" + } + ] + } + ], + "index": 13 + }, + { + "bbox": [ + 313, + 635, + 465, + 647 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 635, + 465, + 647 + ], + "spans": [ + { + "bbox": [ + 313, + 635, + 465, + 647 + ], + "type": "text", + "content": "2.1. Text-to-Image (T2I) models" + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 313, + 653, + 553, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 653, + 553, + 713 + ], + "spans": [ + { + "bbox": [ + 313, + 653, + 553, + 713 + ], + "type": "text", + "content": "Substantial progress has been made on T2I models through the use of diffusion models [20, 59, 60]. These models conceptualize image synthesis as an iterative denoising process guided by textual prompts. The denoising is conditioned on textual embeddings generated by language" + } + ] + } + ], + "index": 15 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "text", + "content": "18380" + } + ] + } + ], + "index": 16 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 1 + }, + { + "para_blocks": [ + { + "bbox": [ + 55, + 72, + 294, + 156 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 72, + 294, + 156 + ], + "spans": [ + { + "bbox": [ + 55, + 72, + 294, + 156 + ], + "type": "text", + "content": "encoders [52, 53] and is performed either in pixel space [2, 43, 54, 57] or latent space [12, 16, 36, 48, 55, 74], followed by cascaded super-resolution [21] or latent-to-image decoding [11] for high-resolution image generation. By training on extensive datasets of image-text pairs [58], these models learn a unified generative space in which visual and linguistic modalities are closely associated." + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 156, + 294, + 205 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 156, + 294, + 205 + ], + "spans": [ + { + "bbox": [ + 55, + 156, + 294, + 205 + ], + "type": "text", + "content": "Nonetheless, current T2I diffusion models still struggle to accurately interpret spatial layout instructions, making it challenging to control the occlusion relationships among multiple objects using textual descriptions alone." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 55, + 213, + 280, + 227 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 213, + 280, + 227 + ], + "spans": [ + { + "bbox": [ + 55, + 213, + 280, + 227 + ], + "type": "text", + "content": "2.2. Layout-guided Text-to-Image (T2I) Models" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 232, + 294, + 267 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 232, + 294, + 267 + ], + "spans": [ + { + "bbox": [ + 55, + 232, + 294, + 267 + ], + "type": "text", + "content": "Layout-guided T2I models aim to generate images that contain objects as specified by the input text description, with their locations aligned with the input layout." + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 268, + 295, + 387 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 268, + 295, + 387 + ], + "spans": [ + { + "bbox": [ + 55, + 268, + 295, + 387 + ], + "type": "text", + "content": "Training-based Models can be categorized into those that fine-tune pre-trained models [1, 7, 10, 23, 31, 51, 72, 76] using the specific input conditions, and those that train additional plug-ins [25, 30, 35, 37, 38, 40, 50, 67, 69, 79] to support various conditions. Some works achieve instance-level control [64, 65, 82, 83] by assigning individual prompts to each object to precisely control their attributes. However, these methods require additional training on large-scale datasets, resulting in significant computational and labor costs." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 388, + 295, + 484 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 388, + 295, + 484 + ], + "spans": [ + { + "bbox": [ + 55, + 388, + 295, + 484 + ], + "type": "text", + "content": "Several methods [22, 26, 71, 81] employ multiple control conditions within a single generation. Sun et al. [63] address the complex relationships among different control conditions by training a multi-control encoder to maintain harmonious results when combining multiple controls. Although these approaches can generate images with harmonious occlusion relationships, they still cannot directly control the occlusion relationship of individual objects." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 485, + 295, + 592 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 485, + 295, + 592 + ], + "spans": [ + { + "bbox": [ + 55, + 485, + 295, + 592 + ], + "type": "text", + "content": "Sampling-based Models often integrate multiple denoising processes, with each process targeting a specific component or region of the image. Liu et al. [33] interpret diffusion models as energy-based models and propose two compositional operators to combine the score functions of different conditions in a single sampling step. MultiDiff. [3] and LRDiff. [49] then propose to utilize shared parameters or constraints to ensure consistency/accuracy of each object in the generated image." + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 55, + 593, + 295, + 641 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 593, + 295, + 641 + ], + "spans": [ + { + "bbox": [ + 55, + 593, + 295, + 641 + ], + "type": "text", + "content": "Although several methods [39, 41] have proposed training-free multimodal control for the T2I generation, they primarily focus on enhancing control techniques without specifically addressing the object occlusion issues." + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 55, + 642, + 295, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 642, + 295, + 713 + ], + "spans": [ + { + "bbox": [ + 55, + 642, + 295, + 713 + ], + "type": "text", + "content": "Attention-based Models can be categorized into forward and backward approaches based on how they employ the attention maps. Forward methods [2, 13, 15, 27, 73] manipulate attention values directly on the feature map. The accuracy of these methods heavily depends on the precision of the layout condition (e.g., segmentation masks). Back-" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 313, + 72, + 553, + 156 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 72, + 553, + 156 + ], + "spans": [ + { + "bbox": [ + 313, + 72, + 553, + 156 + ], + "type": "text", + "content": "ward methods [5, 8, 47, 68, 70, 73] combine the layout condition with the attention map, and utilize an energy function to translate the spatial constraints into the noisy input for iterative updates. These methods also primarily emphasize on positional accuracy while neglecting the occlusion relationships among objects, leading to the unsatisfactory generation of overlapping regions." + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 156, + 553, + 239 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 156, + 553, + 239 + ], + "spans": [ + { + "bbox": [ + 313, + 156, + 553, + 239 + ], + "type": "text", + "content": "In addition, while several customized T2I models [9, 29, 32, 61, 62, 72, 80] attempt to integrate specific input objects into designated regions of the generated image by using additional layout input, they overlook the visibility order among the integrated objects and other objects within or near those regions. This limitation hinders their effectiveness in addressing occlusion issues." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 314, + 250, + 370, + 262 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 250, + 370, + 262 + ], + "spans": [ + { + "bbox": [ + 314, + 250, + 370, + 262 + ], + "type": "text", + "content": "3. Method" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 313, + 270, + 553, + 402 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 270, + 553, + 402 + ], + "spans": [ + { + "bbox": [ + 313, + 270, + 553, + 402 + ], + "type": "text", + "content": "In this work, our goal is to enhance the ability of pre-trained text-to-image (T2I) diffusion models to correctly handle occlusion relationships in multi-object scenarios without training. We note that the overlapping regions of the objects are the main challenge, and they affect the occlusion relationship of the objects in the synthesized image. However, it is difficult for T2I models to understand occlusion relations among objects through text prompts only. In addition, current spatial layouts can only help constrain the spatial location of the objects. They do not indicate the order of objects in the overlapping regions." + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 313, + 402, + 553, + 569 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 402, + 553, + 569 + ], + "spans": [ + { + "bbox": [ + 313, + 402, + 553, + 569 + ], + "type": "text", + "content": "To achieve our goal, we propose a training-free method, VODiff, which introduces object visibility order into the layout-guided T2I generation to effectively address the incorrect object arrangement problem. Fig. 2 shows our framework. It consists of two novel visibility-aware designs: a Sequential Denoising Process (SDP) and a Visibility-Order-Aware (VOA) loss. When generating an image, our SDP strategy divides the whole denoising process into several sub-stages according to the number of user-specified objects and generates the specified objects in each stage sequentially in a bottom-up manner. Meanwhile, at each denoising step, our VOA loss applies dynamic losses to regions with different visibility states, ensuring accurate occlusion relationships among the generated objects." + } + ] + } + ], + "index": 13 + }, + { + "bbox": [ + 313, + 569, + 553, + 617 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 569, + 553, + 617 + ], + "spans": [ + { + "bbox": [ + 313, + 569, + 553, + 617 + ], + "type": "text", + "content": "For the rest of this section, we first briefly introduce the preliminary information of diffusion models and attention layers in Sec. 3.1. We then present how we design the SDP in Sec. 3.2 and detail the VOA loss function in Sec. 3.3." + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 313, + 624, + 400, + 635 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 624, + 400, + 635 + ], + "spans": [ + { + "bbox": [ + 313, + 624, + 400, + 635 + ], + "type": "text", + "content": "3.1. Preliminaries" + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "spans": [ + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "text", + "content": "Diffusion Models [20, 59, 60] learns to approximate a data distribution by progressively denoising a random variable sampled from a unit Gaussian distribution " + }, + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "inline_equation", + "content": "\\mathcal{N}(0,I)" + }, + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "text", + "content": ". At a denoising step " + }, + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "inline_equation", + "content": "t\\in [1,\\dots ,T]" + }, + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "text", + "content": ", the forward-time Markov chain is formulated as " + }, + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "inline_equation", + "content": "x_{t} = \\sqrt{1 - \\beta_{t}} x_{t - 1} + \\sqrt{\\beta_{t}}\\epsilon_{t - 1}" + }, + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "inline_equation", + "content": "\\epsilon \\sim \\mathcal{N}(0,I)" + }, + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "text", + "content": ", and " + }, + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "inline_equation", + "content": "\\beta_{t}" + }, + { + "bbox": [ + 313, + 641, + 553, + 713 + ], + "type": "text", + "content": " is a pre-defined variance schedule." + } + ] + } + ], + "index": 16 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "text", + "content": "18381" + } + ] + } + ], + "index": 17 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 2 + }, + { + "para_blocks": [ + { + "type": "image", + "bbox": [ + 72, + 67, + 541, + 247 + ], + "blocks": [ + { + "bbox": [ + 72, + 67, + 541, + 247 + ], + "lines": [ + { + "bbox": [ + 72, + 67, + 541, + 247 + ], + "spans": [ + { + "bbox": [ + 72, + 67, + 541, + 247 + ], + "type": "image", + "image_path": "2e4359b52be4386cfa615893b5865b4fff7bc1f690515d4dacbb3bb7a3af8615.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 54, + 255, + 555, + 310 + ], + "lines": [ + { + "bbox": [ + 54, + 255, + 555, + 310 + ], + "spans": [ + { + "bbox": [ + 54, + 255, + 555, + 310 + ], + "type": "text", + "content": "Figure 2. Our VODiff incorporates two visibility-aware designs: a sequential denoising process (SDP) and a visibility-order-aware (VOA) loss function. SDP divides the denoising process into several stages according to the visibility order, and performs denoising on different parts sequentially. Meanwhile, the VOA loss is utilized to eliminate the local spatial shifts in the attention layers, further preventing position deviations and incorrect occlusion relationships. Given a text prompt, the spatial layout, and the visibility order of the objects to be synthesized as inputs, our VODiff can generate images that simultaneously adhere to all the constraints." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "image_caption" + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "spans": [ + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "text", + "content": "The reverse-time denoising process can then be defined as: " + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "inline_equation", + "content": "x_{t - 1} = \\tilde{\\alpha}_t x_t + \\tilde{\\beta}_t \\hat{s}_t + \\sigma_t \\epsilon_t" + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "inline_equation", + "content": "\\tilde{\\alpha}_t, \\tilde{\\beta}_t" + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "text", + "content": ", and " + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "inline_equation", + "content": "\\sigma_t" + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "text", + "content": " are calculated coefficients. To generate an image, given a noisy input " + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "inline_equation", + "content": "x_T" + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "text", + "content": " sampled from " + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "inline_equation", + "content": "\\mathcal{N}(0, I)" + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "text", + "content": ", we run denoiser " + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "inline_equation", + "content": "\\epsilon_\\theta" + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "text", + "content": " iteratively to obtain intermediate representations " + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "inline_equation", + "content": "x_{T - 1}, \\dots, x_1, x_0" + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "text", + "content": " where " + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "inline_equation", + "content": "x_0 \\approx x" + }, + { + "bbox": [ + 55, + 317, + 296, + 389 + ], + "type": "text", + "content": " is the final result." + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "spans": [ + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": "Attention Layers are the cornerstone of Stable Diffusion [55]. They used to condition the generation based on the text prompt. A pretrained CLIP encoder [52] is usually utilized to encode the text prompt " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "c = (c_{1}, c_{2}, \\dots, c_{n})" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " is the number of tokens, into text embeddings. These tokens are then used as keys " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "K \\in \\mathbb{R}^{n \\times d}" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " and values " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "V \\in \\mathbb{R}^{n \\times d}" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " in the Cross-Attention layer. The symbol " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "d" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " denotes the feature dimension. For a set of queries " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "Q \\in \\mathbb{R}^{hw \\times d}" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " computed from the feature map of size " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "h \\times w" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": ", the cross-attention map " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "A_{t}" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " at time step " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "t" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " is calculated as: " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "A_{t} = \\text{softmax}\\left(\\frac{QK^{\\top}}{\\sqrt{d}}\\right) \\in [0, 1]^{hw \\times n}" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "A_{t}" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " contains " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " attention maps " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "\\{A_{t}^{1}, \\dots, A_{t}^{n}\\}" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": ", and " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "A_{t}^{i} \\in [0, 1]^{hw}" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " indicates the strength of association between each word token " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "c_{i}" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " and each spatial location on the feature map. A higher attention value in " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "A_{t}^{i}" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " indicates a higher probability of generating the object " + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "inline_equation", + "content": "c_{i}" + }, + { + "bbox": [ + 55, + 390, + 296, + 588 + ], + "type": "text", + "content": " at that spatial location." + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 599, + 246, + 612 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 599, + 246, + 612 + ], + "spans": [ + { + "bbox": [ + 55, + 599, + 246, + 612 + ], + "type": "text", + "content": "3.2. Sequential Denoising Process (SDP)" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 618, + 296, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 618, + 296, + 715 + ], + "spans": [ + { + "bbox": [ + 55, + 618, + 296, + 715 + ], + "type": "text", + "content": "Instead of directly generating all objects on a single layer at once, like what existing works [27, 39, 47] do, we are inspired by the painter's algorithm [42], which sorts all polygons according to their z values and then render them starting from the farthest polygon, and the layer concept in image editing tools such as Photoshop. We propose to divide the generation of the target scene into layers, where each object is assigned a separate layer, and synthesize the ob" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 313, + 317, + 553, + 341 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 317, + 553, + 341 + ], + "spans": [ + { + "bbox": [ + 313, + 317, + 553, + 341 + ], + "type": "text", + "content": "jects sequentially, layer by layer from bottom to top, with the following steps." + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "spans": [ + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": "First, we stratify the target objects into distinct layers according to their visibility order. Specifically, given an input text prompt " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "C" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " containing " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "N" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " target objects; a layout condition represented by bounding boxes " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "B = \\{b_{1}, b_{2}, \\ldots, b_{N}\\}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "b_{n}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " is the bounding box of the " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": "-th target object; and visibility orders " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "V = \\{v_{1}, v_{2}, \\ldots, v_{N}\\}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "v_{n}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " is the visibility order of the " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": "-th target object, we convert the layout condition " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "B" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " into " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "N" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " masks, " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "M = \\{m_{1}, m_{2}, \\ldots, m_{N}\\}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "m_{n}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " is the mask corresponding to the bounding box " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "b_{n}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " with a value of 1 inside the bounding box region and 0 elsewhere. Meanwhile, we extract " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "N" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " object entities from the input text prompt " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "C" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " and denote the set as " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "\\{c_{1}, c_{2}, \\dots, c_{N}\\}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": ". Each bounding box " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "b_{n}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " is associated with an object prompt " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "c_{n}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": ", which describes the content of the target object inside " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "b_{n}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": ". We order the " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "N" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": " target objects in an increasing object visibility order specified by the user, even though there may be non-overlapping objects. The layers from bottom to top can thus be represented as " + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "inline_equation", + "content": "L = \\{(m_{1}, c_{1}, v_{1}), (m_{2}, c_{2}, v_{2}), \\dots, (m_{N}, c_{N}, v_{N})\\}" + }, + { + "bbox": [ + 313, + 342, + 555, + 570 + ], + "type": "text", + "content": "." + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "spans": [ + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": "Second, we divide the reverse-time denoising process (with a total of " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "T" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": " steps) of our SDP into two main parts, " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "[T, T_g]" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "[T_g, 0]" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "T_g" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": " denotes the start time of global denoising. The global denoising is to delineate the overall scene based on the input text prompt " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "C" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": ". The first part of our SDP focuses on per-layer generation based on layers " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "L" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": ". We divide it into " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "N" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": " sub-stages, " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "S = \\{s_1, s_2, \\dots, s_N\\}" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": ", starting from the bottom layer (i.e., the object with the lowest visibility order). Each sub-stage " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "s_n" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": " contains " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "k" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": " denoising steps, ranging from " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "[s_n, s_n + k]" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": ", i.e., " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "T - T_g = kN" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": ". Since the diffusion denoising process is reversed, " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "s_n + k" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": " represents the start time for substage " + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "inline_equation", + "content": "s_n" + }, + { + "bbox": [ + 313, + 570, + 556, + 714 + ], + "type": "text", + "content": ". At each sub-stage, to" + } + ] + } + ], + "index": 8 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "text", + "content": "18382" + } + ] + } + ], + "index": 9 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 3 + }, + { + "para_blocks": [ + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "spans": [ + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "text", + "content": "force " + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "inline_equation", + "content": "c_{n}" + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "text", + "content": " to be generated within the object region, we explicitly incorporate visual guidance into this region. However, we observe that focusing exclusively on a single object " + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "text", + "content": " at each stage can lead to the failure of generating other objects, particularly those whose start time is near " + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "inline_equation", + "content": "T_{g}" + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "text", + "content": ", due to the missing of structure information [2] at later denoising steps. To address this problem, at each stage " + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "text", + "content": ", we apply visual guidance not only for the target object " + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "text", + "content": " but also for other objects " + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "inline_equation", + "content": "j" + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "text", + "content": " (" + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "inline_equation", + "content": "j \\neq n" + }, + { + "bbox": [ + 54, + 72, + 295, + 204 + ], + "type": "text", + "content": "), to ensure the generated layout adheres to the overall requirements. Our scheme for adding visual guidance can be formulated as:" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 124, + 209, + 295, + 221 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 124, + 209, + 295, + 221 + ], + "spans": [ + { + "bbox": [ + 124, + 209, + 295, + 221 + ], + "type": "interline_equation", + "content": "x _ {t, n} = x _ {t} + \\lambda_ {t, n} \\odot m _ {n}, \\tag {1}", + "image_path": "564449ddae860298b2d56e4347586072b89439ec23639b574762331a90e9e47b.jpg" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "spans": [ + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "\\lambda_{t,n}" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": " represents the visual guidance scale of the target object " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": " at time step " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "t" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": ". Note that we allocate a dynamically varying smooth " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "\\lambda_{t,n}" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": " to each object to ensure correct occlusion relationships. Specifically, when processing object " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": " at sub-stage " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": ", the SDP assigns a guidance scale " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "a" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": " to object " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "a/2" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": " to all other objects. As the process transitions from sub-stage " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": " to sub-stage " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "n + 1" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": ", the guidance scale for object " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": " decreases linearly to " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "a/2" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": ", while the guidance scale for object " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "n + 1" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": " increases linearly to " + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "inline_equation", + "content": "a" + }, + { + "bbox": [ + 55, + 224, + 296, + 344 + ], + "type": "text", + "content": ", and guidance scale for the other objects remains unchanged, as:" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 63, + 347, + 295, + 392 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 63, + 347, + 295, + 392 + ], + "spans": [ + { + "bbox": [ + 63, + 347, + 295, + 392 + ], + "type": "interline_equation", + "content": "\\lambda_ {t, n} = \\left\\{ \\begin{array}{l l} a - \\frac {a}{2} \\frac {t - s _ {n - 1}}{k} & \\text {i f} s _ {n - 1} < t < = s _ {n - 1} + k, \\\\ \\frac {a}{2} + \\frac {a}{2} \\frac {t - s _ {n}}{k} & \\text {i f} s _ {n} < t < = s _ {n} + k, \\\\ \\frac {a}{2} & \\text {o t h e r w i s e}, \\end{array} \\right. \\tag {2}", + "image_path": "d249f9d58e5680fb8c1d0ed69383be5b87d868606d5a820ce4063d6a0f7bcd20.jpg" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 393, + 296, + 513 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 393, + 296, + 513 + ], + "spans": [ + { + "bbox": [ + 55, + 393, + 296, + 513 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 55, + 393, + 296, + 513 + ], + "type": "inline_equation", + "content": "a" + }, + { + "bbox": [ + 55, + 393, + 296, + 513 + ], + "type": "text", + "content": " denotes the initial visual guidance scale (refer to Sec. D.4 of the Supp. for more discussion) and " + }, + { + "bbox": [ + 55, + 393, + 296, + 513 + ], + "type": "inline_equation", + "content": "s_0 = T" + }, + { + "bbox": [ + 55, + 393, + 296, + 513 + ], + "type": "text", + "content": ". By assigning a smoothly transitioning visual guidance, we ensure that each sub-stage focuses more on a specific object while maintaining correct occlusion relationships between it and other objects, such that in overlapping regions, objects with a higher visibility order occlude those with a lower one. Finally, we forward the updated noisy input " + }, + { + "bbox": [ + 55, + 393, + 296, + 513 + ], + "type": "inline_equation", + "content": "x_{t,n}" + }, + { + "bbox": [ + 55, + 393, + 296, + 513 + ], + "type": "text", + "content": " to the denoising model " + }, + { + "bbox": [ + 55, + 393, + 296, + 513 + ], + "type": "inline_equation", + "content": "\\epsilon_{\\theta}" + }, + { + "bbox": [ + 55, + 393, + 296, + 513 + ], + "type": "text", + "content": " to estimate noise for each object separately and then combine them together, as:" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 71, + 516, + 295, + 548 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 71, + 516, + 295, + 548 + ], + "spans": [ + { + "bbox": [ + 71, + 516, + 295, + 548 + ], + "type": "interline_equation", + "content": "\\epsilon_ {\\theta} \\left(x _ {t}, t, C\\right) = \\sum_ {n = 1} ^ {N} \\frac {m _ {n}}{\\sum_ {n = 1} ^ {N} m _ {n}} \\otimes \\epsilon_ {\\theta} \\left(x _ {t, n}, t, c _ {n}\\right), \\tag {3}", + "image_path": "cfef6aee7d6ea9296a9bae04c169b09d21ffad5608965fbf4650c39ec2b960fb.jpg" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 552, + 295, + 575 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 552, + 295, + 575 + ], + "spans": [ + { + "bbox": [ + 55, + 552, + 295, + 575 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 55, + 552, + 295, + 575 + ], + "type": "inline_equation", + "content": "\\epsilon_{\\theta}(x_t,t,C)" + }, + { + "bbox": [ + 55, + 552, + 295, + 575 + ], + "type": "text", + "content": " is the estimated noise used to update " + }, + { + "bbox": [ + 55, + 552, + 295, + 575 + ], + "type": "inline_equation", + "content": "x_{t - 1}" + }, + { + "bbox": [ + 55, + 552, + 295, + 575 + ], + "type": "text", + "content": " via the standard diffusion scheduler." + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 55, + 582, + 244, + 596 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 582, + 244, + 596 + ], + "spans": [ + { + "bbox": [ + 55, + 582, + 244, + 596 + ], + "type": "text", + "content": "3.3. Visibility-Order-Aware (VOA) Loss" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 55, + 600, + 296, + 696 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 600, + 296, + 696 + ], + "spans": [ + { + "bbox": [ + 55, + 600, + 296, + 696 + ], + "type": "text", + "content": "Although SDP can help constrain target generation within the specified region using visual guidance, the consecutive conv. layers of the denoising model introduce local shifts, which are then amplified by the attention layers. As a result, adding visual guidance at the input level (i.e., " + }, + { + "bbox": [ + 55, + 600, + 296, + 696 + ], + "type": "inline_equation", + "content": "x_{t}" + }, + { + "bbox": [ + 55, + 600, + 296, + 696 + ], + "type": "text", + "content": ") as a constraint is insufficient and tends to cause object position deviation and false occlusion relationships (see Fig. 5 column 2 for a visual illustration)." + } + ] + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 319, + 72, + 391, + 135 + ], + "blocks": [ + { + "bbox": [ + 319, + 72, + 391, + 135 + ], + "lines": [ + { + "bbox": [ + 319, + 72, + 391, + 135 + ], + "spans": [ + { + "bbox": [ + 319, + 72, + 391, + 135 + ], + "type": "image", + "image_path": "0726e66f0937cde8976af8f2f45d781c5d3838e665730c3ac944c38a82b01c79.jpg" + } + ] + } + ], + "index": 9, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 337, + 137, + 371, + 144 + ], + "lines": [ + { + "bbox": [ + 337, + 137, + 371, + 144 + ], + "spans": [ + { + "bbox": [ + 337, + 137, + 371, + 144 + ], + "type": "text", + "content": "Layout + Order" + } + ] + } + ], + "index": 10, + "angle": 0, + "type": "image_caption" + } + ], + "index": 9 + }, + { + "type": "image", + "bbox": [ + 399, + 72, + 471, + 137 + ], + "blocks": [ + { + "bbox": [ + 399, + 72, + 471, + 137 + ], + "lines": [ + { + "bbox": [ + 399, + 72, + 471, + 137 + ], + "spans": [ + { + "bbox": [ + 399, + 72, + 471, + 137 + ], + "type": "image", + "image_path": "b7b3e19830a58b789a6962629bf870cdb82bf3b59d89d78141f629252401ffba.jpg" + } + ] + } + ], + "index": 11, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 393, + 137, + 477, + 146 + ], + "lines": [ + { + "bbox": [ + 393, + 137, + 477, + 146 + ], + "spans": [ + { + "bbox": [ + 393, + 137, + 477, + 146 + ], + "type": "text", + "content": "moverl mvisible mbackgrd" + } + ] + } + ], + "index": 12, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 313, + 154, + 553, + 198 + ], + "lines": [ + { + "bbox": [ + 313, + 154, + 553, + 198 + ], + "spans": [ + { + "bbox": [ + 313, + 154, + 553, + 198 + ], + "type": "text", + "content": "Figure 3. Region separation illustration in the VOA loss. " + }, + { + "bbox": [ + 313, + 154, + 553, + 198 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{overlap}}" + }, + { + "bbox": [ + 313, + 154, + 553, + 198 + ], + "type": "text", + "content": " represents the regions where the current object " + }, + { + "bbox": [ + 313, + 154, + 553, + 198 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 313, + 154, + 553, + 198 + ], + "type": "text", + "content": " occludes other objects, " + }, + { + "bbox": [ + 313, + 154, + 553, + 198 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{visible}}" + }, + { + "bbox": [ + 313, + 154, + 553, + 198 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 313, + 154, + 553, + 198 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{background}}" + }, + { + "bbox": [ + 313, + 154, + 553, + 198 + ], + "type": "text", + "content": " denote where the current object should and should not be generated in the final image." + } + ] + } + ], + "index": 15, + "angle": 0, + "type": "image_caption" + } + ], + "index": 11 + }, + { + "type": "image", + "bbox": [ + 478, + 72, + 549, + 136 + ], + "blocks": [ + { + "bbox": [ + 478, + 72, + 549, + 136 + ], + "lines": [ + { + "bbox": [ + 478, + 72, + 549, + 136 + ], + "spans": [ + { + "bbox": [ + 478, + 72, + 549, + 136 + ], + "type": "image", + "image_path": "76a7c050cf5dd8b17b460fa1a69427b6309273834ae26436aa2950603ffac531.jpg" + } + ] + } + ], + "index": 13, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 501, + 138, + 530, + 144 + ], + "lines": [ + { + "bbox": [ + 501, + 138, + 530, + 144 + ], + "spans": [ + { + "bbox": [ + 501, + 138, + 530, + 144 + ], + "type": "text", + "content": "Output Image" + } + ] + } + ], + "index": 14, + "angle": 0, + "type": "image_caption" + } + ], + "index": 13 + }, + { + "bbox": [ + 313, + 206, + 554, + 373 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 206, + 554, + 373 + ], + "spans": [ + { + "bbox": [ + 313, + 206, + 554, + 373 + ], + "type": "text", + "content": "To address the above problem, we propose constraining the spatial location and occlusion relationship at feature-level. Specifically, when generating an image with multiple target objects, different object prompts usually have different activations in different regions of the cross-attention map. At a given spatial location, the higher the attention value of a prompt, the more likely the corresponding object will be generated in that region [17]. In addition, when two objects overlap each other, their occlusion relationship is directly influenced by their visibility order. Once the order changes, the content of the overlapping region changes accordingly. Based on this, we transform the spatial layout and occlusion relationship among the objects to the cross-attention map and optimize it for correct image generation." + } + ] + } + ], + "index": 16 + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "spans": [ + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": "As shown in Fig. 3, when processing target object " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ", we first divide the mask " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": " into three parts, overlapping regions " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{overlap}}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ", visible regions " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{visible}}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ", and background regions " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{background}}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ", based on the visibility order and masks of all objects in the layer set " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "L" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ". Specifically, for " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{overlap}}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ", we compute the overlapping regions (i.e., dark orange region) of the current object mask " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": " and the masks of all objects with lower visibility order than the current object (i.e., " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "\\{m_{j}\\}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ", with " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "j \\in [1, n-1]" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ") as " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{overlap}} = \\bigcup_{j=1}^{n-1} m_{n} \\odot m_{j}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ". This is because an object located at an upper layer should obtain a higher attention score in the overlapping regions. For " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{visible}}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ", we first compute the overlapping regions of the current mask " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": " with the masks of all objects that have a higher visibility order (i.e., " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "\\{m_{j}\\}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ", with " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "j \\in [n+1, N]" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": "). We then subtract these overlapping regions from the current object mask, which can be expressed as " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{visible}} = m_{n} - \\bigcup_{j=n+1}^{N} m_{n} \\odot m_{j}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": ". " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{background}}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": " is then obtained by inverting " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{visible}}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": " as " + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "inline_equation", + "content": "m_{n}^{\\mathrm{background}} = 1 - m_{n}^{\\mathrm{visible}}" + }, + { + "bbox": [ + 313, + 374, + 554, + 595 + ], + "type": "text", + "content": "." + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "spans": [ + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "text", + "content": "During the denoising phase for the " + }, + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "text", + "content": "-th target object at time step " + }, + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "inline_equation", + "content": "t" + }, + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "text", + "content": ", as indicated in Eq. 3, we extract the cross-attention maps referred to by the object prompt " + }, + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "inline_equation", + "content": "c_{n}" + }, + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "text", + "content": " from each layer and average them to produce " + }, + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "inline_equation", + "content": "\\mathbf{A}_{t,n}" + }, + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "text", + "content": ". We then compute the loss between the attention map " + }, + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "inline_equation", + "content": "\\mathbf{A}_{t,n}" + }, + { + "bbox": [ + 313, + 595, + 554, + 690 + ], + "type": "text", + "content": " and each of the regions. For the overlapping regions, which are affected by the visibility order, we ensure that objects in the upper layers obtain higher attention scores, as:" + } + ] + } + ], + "index": 18 + }, + { + "bbox": [ + 350, + 700, + 553, + 715 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 350, + 700, + 553, + 715 + ], + "spans": [ + { + "bbox": [ + 350, + 700, + 553, + 715 + ], + "type": "interline_equation", + "content": "\\mathcal {L} _ {t, n} ^ {\\mathrm {o}} = \\alpha_ {o} \\cdot \\left[ 1 - \\min \\left(\\mathbf {A} _ {t, n} \\odot m _ {n} ^ {\\text {o v e r l a p}}\\right) \\right]. \\tag {4}", + "image_path": "b056389bc1224bbaee0c1e1c2945dd4fdbbb701a51b163b1a717829ddb1fba86.jpg" + } + ] + } + ], + "index": 19 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 69, + 702, + 268, + 713 + ], + "type": "page_footnote", + "angle": 0, + "lines": [ + { + "bbox": [ + 69, + 702, + 268, + 713 + ], + "spans": [ + { + "bbox": [ + 69, + 702, + 268, + 713 + ], + "type": "text", + "content": "For simplicity, we omit the classifier-free guidance [19] here." + } + ] + } + ], + "index": 20 + }, + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "text", + "content": "18383" + } + ] + } + ], + "index": 21 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 4 + }, + { + "para_blocks": [ + { + "type": "image", + "bbox": [ + 61, + 68, + 121, + 125 + ], + "blocks": [ + { + "bbox": [ + 61, + 68, + 121, + 125 + ], + "lines": [ + { + "bbox": [ + 61, + 68, + 121, + 125 + ], + "spans": [ + { + "bbox": [ + 61, + 68, + 121, + 125 + ], + "type": "image", + "image_path": "f286221b7c77d1d015e6b94307e065093d084e920aad7a540bd50ee82e31d13a.jpg" + } + ] + } + ], + "index": 6, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 67, + 127, + 188, + 137 + ], + "lines": [ + { + "bbox": [ + 67, + 127, + 188, + 137 + ], + "spans": [ + { + "bbox": [ + 67, + 127, + 188, + 137 + ], + "type": "text", + "content": "\"A picture of a cat, in front of a CD.\"" + } + ] + } + ], + "index": 14, + "angle": 0, + "type": "image_caption" + } + ], + "index": 6 + }, + { + "type": "image", + "bbox": [ + 127, + 68, + 187, + 125 + ], + "blocks": [ + { + "bbox": [ + 127, + 68, + 187, + 125 + ], + "lines": [ + { + "bbox": [ + 127, + 68, + 187, + 125 + ], + "spans": [ + { + "bbox": [ + 127, + 68, + 187, + 125 + ], + "type": "image", + "image_path": "ce17ba2ab414c47029b14a0a20f03afb96287a2ec61845bd18ae535b319a9614.jpg" + } + ] + } + ], + "index": 7, + "angle": 0, + "type": "image_body" + } + ], + "index": 7 + }, + { + "type": "image", + "bbox": [ + 188, + 68, + 245, + 125 + ], + "blocks": [ + { + "bbox": [ + 188, + 68, + 245, + 125 + ], + "lines": [ + { + "bbox": [ + 188, + 68, + 245, + 125 + ], + "spans": [ + { + "bbox": [ + 188, + 68, + 245, + 125 + ], + "type": "image", + "image_path": "9d734215b5e77186a88ff5776f6b8b030ff6c2c7c48e1b2eb44940b5c2ef9eb3.jpg" + } + ] + } + ], + "index": 8, + "angle": 0, + "type": "image_body" + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 246, + 68, + 305, + 125 + ], + "blocks": [ + { + "bbox": [ + 246, + 68, + 305, + 125 + ], + "lines": [ + { + "bbox": [ + 246, + 68, + 305, + 125 + ], + "spans": [ + { + "bbox": [ + 246, + 68, + 305, + 125 + ], + "type": "image", + "image_path": "ae0960715533bf88598d93aa1a4b70a2e58a5b9d23565941ed7ba18916b0665f.jpg" + } + ] + } + ], + "index": 9, + "angle": 0, + "type": "image_body" + } + ], + "index": 9 + }, + { + "type": "image", + "bbox": [ + 307, + 68, + 367, + 125 + ], + "blocks": [ + { + "bbox": [ + 307, + 68, + 367, + 125 + ], + "lines": [ + { + "bbox": [ + 307, + 68, + 367, + 125 + ], + "spans": [ + { + "bbox": [ + 307, + 68, + 367, + 125 + ], + "type": "image", + "image_path": "9df1fea1ee6af4ad2d64997fcbd1a400e5f7a3ca3fb6bd32d23023877c492a3a.jpg" + } + ] + } + ], + "index": 10, + "angle": 0, + "type": "image_body" + } + ], + "index": 10 + }, + { + "type": "image", + "bbox": [ + 370, + 68, + 429, + 125 + ], + "blocks": [ + { + "bbox": [ + 370, + 68, + 429, + 125 + ], + "lines": [ + { + "bbox": [ + 370, + 68, + 429, + 125 + ], + "spans": [ + { + "bbox": [ + 370, + 68, + 429, + 125 + ], + "type": "image", + "image_path": "da0711edb9b22e5fdf281266ad5ec27723e5eceeb5cec78dccb2430f09990a8f.jpg" + } + ] + } + ], + "index": 11, + "angle": 0, + "type": "image_body" + } + ], + "index": 11 + }, + { + "type": "image", + "bbox": [ + 431, + 68, + 490, + 125 + ], + "blocks": [ + { + "bbox": [ + 431, + 68, + 490, + 125 + ], + "lines": [ + { + "bbox": [ + 431, + 68, + 490, + 125 + ], + "spans": [ + { + "bbox": [ + 431, + 68, + 490, + 125 + ], + "type": "image", + "image_path": "b31f5795205c9628c116d0e5ec8c6ccd4c8ae76d331d965d80864c8ea4538fb7.jpg" + } + ] + } + ], + "index": 12, + "angle": 0, + "type": "image_body" + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 492, + 68, + 551, + 125 + ], + "blocks": [ + { + "bbox": [ + 492, + 68, + 551, + 125 + ], + "lines": [ + { + "bbox": [ + 492, + 68, + 551, + 125 + ], + "spans": [ + { + "bbox": [ + 492, + 68, + 551, + 125 + ], + "type": "image", + "image_path": "ba6d297d1203906f1275aa759b9b5eb5e6079fedfe8cbecfe9d52c6cdab61c4e.jpg" + } + ] + } + ], + "index": 13, + "angle": 0, + "type": "image_body" + } + ], + "index": 13 + }, + { + "type": "image", + "bbox": [ + 62, + 139, + 122, + 198 + ], + "blocks": [ + { + "bbox": [ + 62, + 139, + 122, + 198 + ], + "lines": [ + { + "bbox": [ + 62, + 139, + 122, + 198 + ], + "spans": [ + { + "bbox": [ + 62, + 139, + 122, + 198 + ], + "type": "image", + "image_path": "74a4e7daffbb1f5605e43931c895c8613910e365c9a98c008e380c9e4a5ec4f0.jpg" + } + ] + } + ], + "index": 15, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 66, + 200, + 271, + 210 + ], + "lines": [ + { + "bbox": [ + 66, + 200, + 271, + 210 + ], + "spans": [ + { + "bbox": [ + 66, + 200, + 271, + 210 + ], + "type": "text", + "content": "\"A laptop on the table with a plant behind it, and a cup near it.\"" + } + ] + } + ], + "index": 23, + "angle": 0, + "type": "image_caption" + } + ], + "index": 15 + }, + { + "type": "image", + "bbox": [ + 127, + 138, + 187, + 196 + ], + "blocks": [ + { + "bbox": [ + 127, + 138, + 187, + 196 + ], + "lines": [ + { + "bbox": [ + 127, + 138, + 187, + 196 + ], + "spans": [ + { + "bbox": [ + 127, + 138, + 187, + 196 + ], + "type": "image", + "image_path": "063d3786b086993b97a559f2741de26895aec842663364a2495a8ef5dbe6da10.jpg" + } + ] + } + ], + "index": 16, + "angle": 0, + "type": "image_body" + } + ], + "index": 16 + }, + { + "type": "image", + "bbox": [ + 188, + 138, + 245, + 196 + ], + "blocks": [ + { + "bbox": [ + 188, + 138, + 245, + 196 + ], + "lines": [ + { + "bbox": [ + 188, + 138, + 245, + 196 + ], + "spans": [ + { + "bbox": [ + 188, + 138, + 245, + 196 + ], + "type": "image", + "image_path": "563b95716ece1982d690cc18978f54f7343da30aa3588490567b348b7d60b8ed.jpg" + } + ] + } + ], + "index": 17, + "angle": 0, + "type": "image_body" + } + ], + "index": 17 + }, + { + "type": "image", + "bbox": [ + 246, + 138, + 305, + 196 + ], + "blocks": [ + { + "bbox": [ + 246, + 138, + 305, + 196 + ], + "lines": [ + { + "bbox": [ + 246, + 138, + 305, + 196 + ], + "spans": [ + { + "bbox": [ + 246, + 138, + 305, + 196 + ], + "type": "image", + "image_path": "0ee7477788993e3024b0a376df272a5d9cef0127487f94255cadb7092d74d8ba.jpg" + } + ] + } + ], + "index": 18, + "angle": 0, + "type": "image_body" + } + ], + "index": 18 + }, + { + "type": "image", + "bbox": [ + 307, + 138, + 366, + 196 + ], + "blocks": [ + { + "bbox": [ + 307, + 138, + 366, + 196 + ], + "lines": [ + { + "bbox": [ + 307, + 138, + 366, + 196 + ], + "spans": [ + { + "bbox": [ + 307, + 138, + 366, + 196 + ], + "type": "image", + "image_path": "be4407ad3e94c30cddaea4eda069f97ac82efc83503b3d7b739568a267444da5.jpg" + } + ] + } + ], + "index": 19, + "angle": 0, + "type": "image_body" + } + ], + "index": 19 + }, + { + "type": "image", + "bbox": [ + 370, + 138, + 429, + 196 + ], + "blocks": [ + { + "bbox": [ + 370, + 138, + 429, + 196 + ], + "lines": [ + { + "bbox": [ + 370, + 138, + 429, + 196 + ], + "spans": [ + { + "bbox": [ + 370, + 138, + 429, + 196 + ], + "type": "image", + "image_path": "2edd5e00bffc71990355a2151cd9a70192563c824b967ef62ed8d8e5878a8756.jpg" + } + ] + } + ], + "index": 20, + "angle": 0, + "type": "image_body" + } + ], + "index": 20 + }, + { + "type": "image", + "bbox": [ + 431, + 138, + 490, + 196 + ], + "blocks": [ + { + "bbox": [ + 431, + 138, + 490, + 196 + ], + "lines": [ + { + "bbox": [ + 431, + 138, + 490, + 196 + ], + "spans": [ + { + "bbox": [ + 431, + 138, + 490, + 196 + ], + "type": "image", + "image_path": "21dd74696ef2ff62afd698794830001a938fa8943442d8ec0f388e11b3a7fcf0.jpg" + } + ] + } + ], + "index": 21, + "angle": 0, + "type": "image_body" + } + ], + "index": 21 + }, + { + "type": "image", + "bbox": [ + 492, + 138, + 551, + 196 + ], + "blocks": [ + { + "bbox": [ + 492, + 138, + 551, + 196 + ], + "lines": [ + { + "bbox": [ + 492, + 138, + 551, + 196 + ], + "spans": [ + { + "bbox": [ + 492, + 138, + 551, + 196 + ], + "type": "image", + "image_path": "7e3083d800d017630758cd36e8de6ab22b08836b478e4a84ee7f9aff1ab6c428.jpg" + } + ] + } + ], + "index": 22, + "angle": 0, + "type": "image_body" + } + ], + "index": 22 + }, + { + "type": "image", + "bbox": [ + 62, + 212, + 121, + 270 + ], + "blocks": [ + { + "bbox": [ + 62, + 212, + 121, + 270 + ], + "lines": [ + { + "bbox": [ + 62, + 212, + 121, + 270 + ], + "spans": [ + { + "bbox": [ + 62, + 212, + 121, + 270 + ], + "type": "image", + "image_path": "b280df6c33bf337eb85bf78871cc45a18bbc7e12810b012de9d1b3af6746ca54.jpg" + } + ] + } + ], + "index": 24, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 65, + 273, + 297, + 283 + ], + "lines": [ + { + "bbox": [ + 65, + 273, + 297, + 283 + ], + "spans": [ + { + "bbox": [ + 65, + 273, + 297, + 283 + ], + "type": "text", + "content": "\"Left chair and right chair, a lamp and a plant in front of a bookshelf.\"" + } + ] + } + ], + "index": 32, + "angle": 0, + "type": "image_caption" + } + ], + "index": 24 + }, + { + "type": "image", + "bbox": [ + 127, + 213, + 187, + 269 + ], + "blocks": [ + { + "bbox": [ + 127, + 213, + 187, + 269 + ], + "lines": [ + { + "bbox": [ + 127, + 213, + 187, + 269 + ], + "spans": [ + { + "bbox": [ + 127, + 213, + 187, + 269 + ], + "type": "image", + "image_path": "582497669611b385566a15c4e9b3e0e11549314bd54777fb6575cb48db381b10.jpg" + } + ] + } + ], + "index": 25, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 102, + 297, + 506, + 308 + ], + "lines": [ + { + "bbox": [ + 102, + 297, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 102, + 297, + 506, + 308 + ], + "type": "text", + "content": "Figure 4. Qualitative comparison with six SOTA methods in Tab. 1. Training-based methods are labeled with " + }, + { + "bbox": [ + 102, + 297, + 506, + 308 + ], + "type": "inline_equation", + "content": "\\dagger" + }, + { + "bbox": [ + 102, + 297, + 506, + 308 + ], + "type": "text", + "content": "." + } + ] + } + ], + "index": 33, + "angle": 0, + "type": "image_caption" + } + ], + "index": 25 + }, + { + "type": "image", + "bbox": [ + 188, + 213, + 245, + 269 + ], + "blocks": [ + { + "bbox": [ + 188, + 213, + 245, + 269 + ], + "lines": [ + { + "bbox": [ + 188, + 213, + 245, + 269 + ], + "spans": [ + { + "bbox": [ + 188, + 213, + 245, + 269 + ], + "type": "image", + "image_path": "657650ba2ba1848ac104b557bea57d507980cc91ad17e2655f20a7914cc106f4.jpg" + } + ] + } + ], + "index": 26, + "angle": 0, + "type": "image_body" + } + ], + "index": 26 + }, + { + "type": "image", + "bbox": [ + 246, + 213, + 305, + 269 + ], + "blocks": [ + { + "bbox": [ + 246, + 213, + 305, + 269 + ], + "lines": [ + { + "bbox": [ + 246, + 213, + 305, + 269 + ], + "spans": [ + { + "bbox": [ + 246, + 213, + 305, + 269 + ], + "type": "image", + "image_path": "3d2e2486f491fba8ad002fc48000a214a3012963043f68fa76f9ec785fba5d19.jpg" + } + ] + } + ], + "index": 27, + "angle": 0, + "type": "image_body" + } + ], + "index": 27 + }, + { + "type": "image", + "bbox": [ + 307, + 213, + 367, + 269 + ], + "blocks": [ + { + "bbox": [ + 307, + 213, + 367, + 269 + ], + "lines": [ + { + "bbox": [ + 307, + 213, + 367, + 269 + ], + "spans": [ + { + "bbox": [ + 307, + 213, + 367, + 269 + ], + "type": "image", + "image_path": "afeaeed0f2717ad1f2d0133eb34d2c20d8ed2957758e4817b3180e49a6e86e4b.jpg" + } + ] + } + ], + "index": 28, + "angle": 0, + "type": "image_body" + } + ], + "index": 28 + }, + { + "type": "image", + "bbox": [ + 370, + 213, + 429, + 269 + ], + "blocks": [ + { + "bbox": [ + 370, + 213, + 429, + 269 + ], + "lines": [ + { + "bbox": [ + 370, + 213, + 429, + 269 + ], + "spans": [ + { + "bbox": [ + 370, + 213, + 429, + 269 + ], + "type": "image", + "image_path": "2613240aa8c6356dd8c6dc2ff6cc69d524b1762a3f90e9205cd64cddc8d06d93.jpg" + } + ] + } + ], + "index": 29, + "angle": 0, + "type": "image_body" + } + ], + "index": 29 + }, + { + "type": "image", + "bbox": [ + 431, + 213, + 490, + 269 + ], + "blocks": [ + { + "bbox": [ + 431, + 213, + 490, + 269 + ], + "lines": [ + { + "bbox": [ + 431, + 213, + 490, + 269 + ], + "spans": [ + { + "bbox": [ + 431, + 213, + 490, + 269 + ], + "type": "image", + "image_path": "d91a5301d1caf3d62be6c3decee53a5b22e71be91798fc9d6818783dc72f24ba.jpg" + } + ] + } + ], + "index": 30, + "angle": 0, + "type": "image_body" + } + ], + "index": 30 + }, + { + "type": "image", + "bbox": [ + 492, + 213, + 551, + 269 + ], + "blocks": [ + { + "bbox": [ + 492, + 213, + 551, + 269 + ], + "lines": [ + { + "bbox": [ + 492, + 213, + 551, + 269 + ], + "spans": [ + { + "bbox": [ + 492, + 213, + 551, + 269 + ], + "type": "image", + "image_path": "e1c8306413edda32a103e8f9d238359270439bbeb8079cf3253c0cb0a13dbe7e.jpg" + } + ] + } + ], + "index": 31, + "angle": 0, + "type": "image_body" + } + ], + "index": 31 + }, + { + "bbox": [ + 55, + 318, + 296, + 367 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 318, + 296, + 367 + ], + "spans": [ + { + "bbox": [ + 55, + 318, + 296, + 367 + ], + "type": "text", + "content": "We then encourage target object " + }, + { + "bbox": [ + 55, + 318, + 296, + 367 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 55, + 318, + 296, + 367 + ], + "type": "text", + "content": " to be generated within the visible regions by increasing the attention score through " + }, + { + "bbox": [ + 55, + 318, + 296, + 367 + ], + "type": "inline_equation", + "content": "\\mathcal{L}_{t,n}^{\\mathrm{v}}" + }, + { + "bbox": [ + 55, + 318, + 296, + 367 + ], + "type": "text", + "content": ", and prevent the target deviation by weakening the attention score of the background regions via " + }, + { + "bbox": [ + 55, + 318, + 296, + 367 + ], + "type": "inline_equation", + "content": "\\mathcal{L}_{t,n}^{\\mathrm{b}}" + }, + { + "bbox": [ + 55, + 318, + 296, + 367 + ], + "type": "text", + "content": ":" + } + ] + } + ], + "index": 34 + }, + { + "bbox": [ + 88, + 373, + 295, + 388 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 88, + 373, + 295, + 388 + ], + "spans": [ + { + "bbox": [ + 88, + 373, + 295, + 388 + ], + "type": "interline_equation", + "content": "\\mathcal {L} _ {t, n} ^ {\\mathrm {v}} = \\alpha_ {v i s} \\cdot \\left[ 1 - \\max \\left(\\mathbf {A} _ {t, n} \\odot m _ {n} ^ {\\text {v i s i b l e}}\\right) \\right], \\tag {5}", + "image_path": "208e478fd0a6bd9a9ff35adca49e153671695c13998a654e3ae4593a12bec28d.jpg" + } + ] + } + ], + "index": 35 + }, + { + "bbox": [ + 95, + 389, + 295, + 404 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 95, + 389, + 295, + 404 + ], + "spans": [ + { + "bbox": [ + 95, + 389, + 295, + 404 + ], + "type": "interline_equation", + "content": "\\mathcal {L} _ {t, n} ^ {\\mathrm {b}} = \\alpha_ {b a c} \\cdot \\max \\left(\\mathbf {A} _ {t, n} \\odot m _ {n} ^ {\\text {b a c k g r o u n d}}\\right), \\tag {6}", + "image_path": "d4b50cd8e4a4be28bef92d595e17c59f82dd943d4601128c9138864179c57b7b.jpg" + } + ] + } + ], + "index": 36 + }, + { + "bbox": [ + 55, + 407, + 295, + 430 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 407, + 295, + 430 + ], + "spans": [ + { + "bbox": [ + 55, + 407, + 295, + 430 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 55, + 407, + 295, + 430 + ], + "type": "inline_equation", + "content": "\\alpha_{o}" + }, + { + "bbox": [ + 55, + 407, + 295, + 430 + ], + "type": "text", + "content": ", " + }, + { + "bbox": [ + 55, + 407, + 295, + 430 + ], + "type": "inline_equation", + "content": "\\alpha_{vis}" + }, + { + "bbox": [ + 55, + 407, + 295, + 430 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 55, + 407, + 295, + 430 + ], + "type": "inline_equation", + "content": "\\alpha_{bac}" + }, + { + "bbox": [ + 55, + 407, + 295, + 430 + ], + "type": "text", + "content": " are coefficients for the three loss functions. Finally, the total loss is:" + } + ] + } + ], + "index": 37 + }, + { + "bbox": [ + 112, + 430, + 294, + 458 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 112, + 430, + 294, + 458 + ], + "spans": [ + { + "bbox": [ + 112, + 430, + 294, + 458 + ], + "type": "interline_equation", + "content": "\\mathcal {L} _ {t} = \\sum_ {n} ^ {N} \\left(\\mathcal {L} _ {t, n} ^ {\\mathrm {o}} + \\mathcal {L} _ {t, n} ^ {\\mathrm {v}} + \\mathcal {L} _ {t, n} ^ {\\mathrm {b}}\\right). \\tag {7}", + "image_path": "18c0b87d4da0a0a0a342104f7a36161d7d855dac004f5a4a28635fcaa8241097.jpg" + } + ] + } + ], + "index": 38 + }, + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "spans": [ + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "text", + "content": "With the visibility-order-aware loss, we modify the noise sample " + }, + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "inline_equation", + "content": "x_{t}" + }, + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "text", + "content": " at each denoising step to minimize the loss using gradient descent " + }, + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "inline_equation", + "content": "\\hat{x}_{t} \\gets x_{t} - \\alpha \\nabla_{x_{t}}\\mathcal{L}_{t}" + }, + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "inline_equation", + "content": "\\alpha" + }, + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "text", + "content": " controls the influence of the optimization in the denoising process. In each step " + }, + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "inline_equation", + "content": "t" + }, + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "text", + "content": ", we iterate the gradient descent process " + }, + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "inline_equation", + "content": "\\tau" + }, + { + "bbox": [ + 55, + 458, + 296, + 553 + ], + "type": "text", + "content": " times. By applying the VOA loss on all objects at each sub-stage, VODiff alleviates the guidance deviation and achieves more accurate object spatial position and occlusions." + } + ] + } + ], + "index": 39 + }, + { + "bbox": [ + 55, + 557, + 131, + 570 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 557, + 131, + 570 + ], + "spans": [ + { + "bbox": [ + 55, + 557, + 131, + 570 + ], + "type": "text", + "content": "4. Experiment" + } + ] + } + ], + "index": 40 + }, + { + "bbox": [ + 55, + 576, + 115, + 588 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 576, + 115, + 588 + ], + "spans": [ + { + "bbox": [ + 55, + 576, + 115, + 588 + ], + "type": "text", + "content": "4.1. Metrics" + } + ] + } + ], + "index": 41 + }, + { + "bbox": [ + 55, + 594, + 244, + 605 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 594, + 244, + 605 + ], + "spans": [ + { + "bbox": [ + 55, + 594, + 244, + 605 + ], + "type": "text", + "content": "We adopt a few metrics to evaluate our model:" + } + ] + } + ], + "index": 42 + }, + { + "bbox": [ + 55, + 605, + 296, + 713 + ], + "type": "list", + "angle": 0, + "index": 45, + "blocks": [ + { + "bbox": [ + 55, + 605, + 295, + 665 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 605, + 295, + 665 + ], + "spans": [ + { + "bbox": [ + 55, + 605, + 295, + 665 + ], + "type": "text", + "content": "- FID: We use Fréchet Inception Distance [18] to evaluate the image quality. It compares the distribution of generated images to that of real images. A lower FID score indicates a higher image quality, reflecting closer alignment with real images on visual features and diversity." + } + ] + } + ], + "index": 43 + }, + { + "bbox": [ + 55, + 665, + 296, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 665, + 296, + 713 + ], + "spans": [ + { + "bbox": [ + 55, + 665, + 296, + 713 + ], + "type": "text", + "content": "- CLIP-Score: We use the CLIP-Score to compute the distance between input textual features and generated image features, evaluating the fidelity of the generated image to the textual input." + } + ] + } + ], + "index": 44 + } + ], + "sub_type": "text" + }, + { + "bbox": [ + 314, + 318, + 554, + 475 + ], + "type": "list", + "angle": 0, + "index": 48, + "blocks": [ + { + "bbox": [ + 314, + 318, + 554, + 377 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 318, + 554, + 377 + ], + "spans": [ + { + "bbox": [ + 314, + 318, + 554, + 377 + ], + "type": "text", + "content": "- **Layout Alignment (LA):** We use Grounding-DINO [34] to assess Layout alignment accuracy. It detects bounding boxes for each instance and then compute the mean IoU between the detected and Ground Truth boxes. The box threshold is set to 0.5 to avoid incorrect box detection." + } + ] + } + ], + "index": 46 + }, + { + "bbox": [ + 314, + 378, + 554, + 475 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 378, + 554, + 475 + ], + "spans": [ + { + "bbox": [ + 314, + 378, + 554, + 475 + ], + "type": "text", + "content": "- Occlusion Alignment (OA): We employ GPT-4V to determine the occlusion relationships between objects in the generated images by prompting it with the image and the object names contained in the image. We then compare the visual order returned by GPT-4V with the ground truth visibility order. If the two match with each other, we count it as accurate. Finally, we compute the percentage of accurate occlusion relationships from all the samples." + } + ] + } + ], + "index": 47 + } + ], + "sub_type": "text" + }, + { + "bbox": [ + 313, + 481, + 382, + 493 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 481, + 382, + 493 + ], + "spans": [ + { + "bbox": [ + 313, + 481, + 382, + 493 + ], + "type": "text", + "content": "4.2.VObench" + } + ] + } + ], + "index": 49 + }, + { + "bbox": [ + 313, + 496, + 555, + 651 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 496, + 555, + 651 + ], + "spans": [ + { + "bbox": [ + 313, + 496, + 555, + 651 + ], + "type": "text", + "content": "Since there are no visibility-order-aware datasets available for evaluation, we curate a VOBench for this purpose. All the raw image data in VOBench are manually collected from the Internet. The data annotation process is as follows. We first employ GroundingDINO [34] and GPT-4V to generate the bounding boxes and text prompts for each object in the image, and manually assign the visibility order to each object box. Based on the number of boxes detected by GroundingDINO, we then divide the images into four categories containing 2-5 objects. For each category, we manually select 50 images based on the degree of overlap between boxes to make sure that various occlusion relations are in our VOBench." + } + ] + } + ], + "index": 50 + }, + { + "bbox": [ + 313, + 659, + 539, + 672 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 659, + 539, + 672 + ], + "spans": [ + { + "bbox": [ + 313, + 659, + 539, + 672 + ], + "type": "text", + "content": "4.3. Comparison with State-of-the-Art Methods" + } + ] + } + ], + "index": 51 + }, + { + "bbox": [ + 313, + 677, + 554, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 677, + 554, + 713 + ], + "spans": [ + { + "bbox": [ + 313, + 677, + 554, + 713 + ], + "type": "text", + "content": "We compare VODiff with 12 layout-guided text-to-image generation methods, including seven training-free methods: BoxDiff [70], R&B [68], Attention-Refocusing [47]," + } + ] + } + ], + "index": 52 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 66, + 53, + 119, + 64 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 66, + 53, + 119, + 64 + ], + "spans": [ + { + "bbox": [ + 66, + 53, + 119, + 64 + ], + "type": "text", + "content": "(a) Layout+Order" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 123, + 53, + 308, + 65 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 123, + 53, + 308, + 65 + ], + "spans": [ + { + "bbox": [ + 123, + 53, + 308, + 65 + ], + "type": "text", + "content": "(b) InstanceDiff.† [65] (c) AnyControl† [63](d) ControlNet† [79]" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 316, + 53, + 364, + 64 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 53, + 364, + 64 + ], + "spans": [ + { + "bbox": [ + 316, + 53, + 364, + 64 + ], + "type": "text", + "content": "(e) MIGC† [83]" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 373, + 54, + 426, + 64 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 373, + 54, + 426, + 64 + ], + "spans": [ + { + "bbox": [ + 373, + 54, + 426, + 64 + ], + "type": "text", + "content": "(f) RPG-Diff. [73]" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 433, + 54, + 492, + 64 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 433, + 54, + 492, + 64 + ], + "spans": [ + { + "bbox": [ + 433, + 54, + 492, + 64 + ], + "type": "text", + "content": "(g) FreeControl [39]" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 507, + 54, + 533, + 64 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 507, + 54, + 533, + 64 + ], + "spans": [ + { + "bbox": [ + 507, + 54, + 533, + 64 + ], + "type": "text", + "content": "(h) Ours" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "text", + "content": "18384" + } + ] + } + ], + "index": 53 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 5 + }, + { + "para_blocks": [ + { + "type": "table", + "bbox": [ + 60, + 107, + 547, + 192 + ], + "blocks": [ + { + "bbox": [ + 55, + 70, + 555, + 105 + ], + "lines": [ + { + "bbox": [ + 55, + 70, + 555, + 105 + ], + "spans": [ + { + "bbox": [ + 55, + 70, + 555, + 105 + ], + "type": "text", + "content": "Table 1. Quantitative comparison of our VODiff with training-based and training-free methods. LA and OA represent the layout and occlusion alignment accuracy, respectively. AR denotes the average ranking from the user study, while Q, L, and O indicate that the ranking is based on Quality, Layout, and Occlusion. The best and second-best performances are marked in bold and underlined, respectively." + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "table_caption" + }, + { + "bbox": [ + 60, + 107, + 547, + 192 + ], + "lines": [ + { + "bbox": [ + 60, + 107, + 547, + 192 + ], + "spans": [ + { + "bbox": [ + 60, + 107, + 547, + 192 + ], + "type": "table", + "html": "
MetricsTraining-basedTraining-free
SmartCtrl [35]ControlNet [79]MIGC[83]InstanceDiff. [65]AnyControl [63]BoxDiff [70]R&B [68]AR [47]MultiDiff. [3]DenseDiff. [27]FreeCtrl [39]RPG [73]Ours
FID ↓10.5818.919.8210.2110.3521.9912.3714.8716.3421.6817.3511.6610.03
CLIP-Score ↑29.4529.3429.6129.5329.6329.4628.7229.5927.4229.3329.1229.5729.73
LA ↑54.8152.7354.6253.5854.7713.5030.8829.7916.6024.1136.1451.1555.11
OA ↑55.0051.5065.0063.5069.5018.5034.5032.0021.5024.0028.5051.5082.50
AR-Q ↓5.3312.132.476.203.058.1510.054.1312.687.109.005.121.50
AR-L ↓2.166.214.045.113.1212.989.0210.0712.0311.148.087.171.73
AR-O ↓6.255.123.184.022.0512.749.0110.1112.0411.098.037.101.26
", + "image_path": "5c3d44859523e0d83c5ee910e436a5db305ace238c3fe3c72931aa73778a3dcc.jpg" + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "table_body" + } + ], + "index": 1 + }, + { + "type": "table", + "bbox": [ + 64, + 228, + 284, + 317 + ], + "blocks": [ + { + "bbox": [ + 55, + 201, + 295, + 224 + ], + "lines": [ + { + "bbox": [ + 55, + 201, + 295, + 224 + ], + "spans": [ + { + "bbox": [ + 55, + 201, + 295, + 224 + ], + "type": "text", + "content": "Table 2. Ablation study on components. Constant and Dynamic represent the state of change of guidance scale in the SDP." + } + ] + } + ], + "index": 2, + "angle": 0, + "type": "table_caption" + }, + { + "bbox": [ + 64, + 228, + 284, + 317 + ], + "lines": [ + { + "bbox": [ + 64, + 228, + 284, + 317 + ], + "spans": [ + { + "bbox": [ + 64, + 228, + 284, + 317 + ], + "type": "table", + "html": "
SDP \nConstant DynamicVOAFID ↓Clip-Score ↑LA ↑OA ↑
✓9.9827.8532.3354.50
✓13.3429.0143.2573.50
✓✓11.6829.3343.3669.50
✓✓10.0329.7355.1182.50
", + "image_path": "4f63dad8f4729f4a8ba2d36b69ab0962d834741367c7df12bf9bcedef1b286e6.jpg" + } + ] + } + ], + "index": 3, + "angle": 0, + "type": "table_body" + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 331, + 296, + 415 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 331, + 296, + 415 + ], + "spans": [ + { + "bbox": [ + 55, + 331, + 296, + 415 + ], + "type": "text", + "content": "MultiDiffusion [3], DenseDiffusion [27], FreeControl [39] and RPG-Diffusion [73], and five training-based methods: SmartControl [35], ControlNet [79], AnyControl [63], MIGC [83] and InstanceDiffusion [65]. Since these methods do not support visibility order as an input, we include descriptions of object occlusion relationships in the text prompts to ensure fairness." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 420, + 296, + 552 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 420, + 296, + 552 + ], + "spans": [ + { + "bbox": [ + 55, + 420, + 296, + 552 + ], + "type": "text", + "content": "Qualitative Comparison. As shown in Fig. 4, we visually compare VODiff with six top-performing methods. The images generated by VODiff exhibit superior spatial consistency and occlusion handling. Specifically, our method handles varying numbers of objects and degrees of occlusion remarkably well. In contrast, other methods exhibit issues such as incorrect spatial positioning (e.g., the 2nd row, (b-f)) or missing objects (e.g., the 3rd row, (d-f)). In addition, although some methods manage to generate correct spatial relationships, they fail in occlusion handling (e.g., the 1st row, (b,e,f,g))." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 558, + 296, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 558, + 296, + 713 + ], + "spans": [ + { + "bbox": [ + 55, + 558, + 296, + 713 + ], + "type": "text", + "content": "Quantitative Comparison. As shown in Tab. 1, we evaluate our method against both training-free and training-based approaches. Our method significantly outperforms the training-free methods across all metrics. We achieve higher CLIP-Score (29.73), LA (55.11), and OA (82.50) scores, indicating superior image quality, layout alignment, and occlusion handling. Despite being training-free, our method outperforms training-based methods in six out of seven metrics. We achieve higher CLIP-Score (29.73 vs 29.63 by AnyControl), LA (55.11 vs 54.81 by SmartControl), and OA (82.50 vs 69.50 by AnyControl) scores. This demonstrates the effectiveness of our method in generating text-/layout-aligned and occlusion handled images without" + } + ] + } + ], + "index": 6 + }, + { + "type": "image", + "bbox": [ + 323, + 217, + 369, + 255 + ], + "blocks": [ + { + "bbox": [ + 323, + 217, + 369, + 255 + ], + "lines": [ + { + "bbox": [ + 323, + 217, + 369, + 255 + ], + "spans": [ + { + "bbox": [ + 323, + 217, + 369, + 255 + ], + "type": "image", + "image_path": "cecbb91620f2c77915d98270bf5fde2d7c24e901ab3f4010a80789567c45d16f.jpg" + } + ] + } + ], + "index": 8, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 329, + 256, + 460, + 265 + ], + "lines": [ + { + "bbox": [ + 329, + 256, + 460, + 265 + ], + "spans": [ + { + "bbox": [ + 329, + 256, + 460, + 265 + ], + "type": "text", + "content": "\"A bowl of strawberries and a cup on a blanket\"" + } + ] + } + ], + "index": 17, + "angle": 0, + "type": "image_caption" + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 370, + 215, + 412, + 255 + ], + "blocks": [ + { + "bbox": [ + 323, + 205, + 368, + 214 + ], + "lines": [ + { + "bbox": [ + 323, + 205, + 368, + 214 + ], + "spans": [ + { + "bbox": [ + 323, + 205, + 368, + 214 + ], + "type": "text", + "content": "Layout + Order" + } + ] + } + ], + "index": 7, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 386, + 205, + 400, + 213 + ], + "lines": [ + { + "bbox": [ + 386, + 205, + 400, + 213 + ], + "spans": [ + { + "bbox": [ + 386, + 205, + 400, + 213 + ], + "type": "text", + "content": "SDP" + } + ] + } + ], + "index": 9, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 370, + 215, + 412, + 255 + ], + "lines": [ + { + "bbox": [ + 370, + 215, + 412, + 255 + ], + "spans": [ + { + "bbox": [ + 370, + 215, + 412, + 255 + ], + "type": "image", + "image_path": "41c8934317f3065450f9d280f8e83029236964029378cae0582e953881982acd.jpg" + } + ] + } + ], + "index": 10, + "angle": 0, + "type": "image_body" + } + ], + "index": 10 + }, + { + "type": "image", + "bbox": [ + 414, + 215, + 457, + 255 + ], + "blocks": [ + { + "bbox": [ + 427, + 205, + 444, + 213 + ], + "lines": [ + { + "bbox": [ + 427, + 205, + 444, + 213 + ], + "spans": [ + { + "bbox": [ + 427, + 205, + 444, + 213 + ], + "type": "text", + "content": "VOA" + } + ] + } + ], + "index": 11, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 414, + 215, + 457, + 255 + ], + "lines": [ + { + "bbox": [ + 414, + 215, + 457, + 255 + ], + "spans": [ + { + "bbox": [ + 414, + 215, + 457, + 255 + ], + "type": "image", + "image_path": "f8d3e1200333f75876254ebc53169bda58386ae479f89ca832f206c1de7f6068.jpg" + } + ] + } + ], + "index": 12, + "angle": 0, + "type": "image_body" + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 459, + 215, + 501, + 255 + ], + "blocks": [ + { + "bbox": [ + 463, + 205, + 497, + 213 + ], + "lines": [ + { + "bbox": [ + 463, + 205, + 497, + 213 + ], + "spans": [ + { + "bbox": [ + 463, + 205, + 497, + 213 + ], + "type": "text", + "content": "VOA+SDPc" + } + ] + } + ], + "index": 13, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 459, + 215, + 501, + 255 + ], + "lines": [ + { + "bbox": [ + 459, + 215, + 501, + 255 + ], + "spans": [ + { + "bbox": [ + 459, + 215, + 501, + 255 + ], + "type": "image", + "image_path": "3bac4bcb48d2485fd1ec6048dd06ef7ea760667f5fd8498758ef20b9962b4110.jpg" + } + ] + } + ], + "index": 14, + "angle": 0, + "type": "image_body" + } + ], + "index": 14 + }, + { + "type": "image", + "bbox": [ + 502, + 215, + 545, + 255 + ], + "blocks": [ + { + "bbox": [ + 517, + 205, + 531, + 213 + ], + "lines": [ + { + "bbox": [ + 517, + 205, + 531, + 213 + ], + "spans": [ + { + "bbox": [ + 517, + 205, + 531, + 213 + ], + "type": "text", + "content": "Ours" + } + ] + } + ], + "index": 15, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 502, + 215, + 545, + 255 + ], + "lines": [ + { + "bbox": [ + 502, + 215, + 545, + 255 + ], + "spans": [ + { + "bbox": [ + 502, + 215, + 545, + 255 + ], + "type": "image", + "image_path": "2e8af02634b88617c21adff2ce9baaeb55674dacddc9f187f1cd5a5d25c3caab.jpg" + } + ] + } + ], + "index": 16, + "angle": 0, + "type": "image_body" + } + ], + "index": 16 + }, + { + "type": "image", + "bbox": [ + 323, + 266, + 367, + 306 + ], + "blocks": [ + { + "bbox": [ + 323, + 266, + 367, + 306 + ], + "lines": [ + { + "bbox": [ + 323, + 266, + 367, + 306 + ], + "spans": [ + { + "bbox": [ + 323, + 266, + 367, + 306 + ], + "type": "image", + "image_path": "03374d504f92570bcf9c0141629f3e33f3305ce8cf55a682808345a9935959b1.jpg" + } + ] + } + ], + "index": 18, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 329, + 308, + 428, + 316 + ], + "lines": [ + { + "bbox": [ + 329, + 308, + 428, + 316 + ], + "spans": [ + { + "bbox": [ + 329, + 308, + 428, + 316 + ], + "type": "text", + "content": "\"A man and a dog in front of a sofa\"" + } + ] + } + ], + "index": 23, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 313, + 319, + 554, + 353 + ], + "lines": [ + { + "bbox": [ + 313, + 319, + 554, + 353 + ], + "spans": [ + { + "bbox": [ + 313, + 319, + 554, + 353 + ], + "type": "text", + "content": "Figure 5. Visual comparison of the ablation study in Tab. 2. " + }, + { + "bbox": [ + 313, + 319, + 554, + 353 + ], + "type": "inline_equation", + "content": "\\mathrm{SDP}_c" + }, + { + "bbox": [ + 313, + 319, + 554, + 353 + ], + "type": "text", + "content": " denotes that the constant guidance scale is used. By default, the dynamic guidance scale is used in SDP." + } + ] + } + ], + "index": 24, + "angle": 0, + "type": "image_caption" + } + ], + "index": 18 + }, + { + "type": "image", + "bbox": [ + 370, + 266, + 412, + 306 + ], + "blocks": [ + { + "bbox": [ + 370, + 266, + 412, + 306 + ], + "lines": [ + { + "bbox": [ + 370, + 266, + 412, + 306 + ], + "spans": [ + { + "bbox": [ + 370, + 266, + 412, + 306 + ], + "type": "image", + "image_path": "299d32385b9527f212ba2dea35a6140e9074ea35062a227b4059664b609b1b09.jpg" + } + ] + } + ], + "index": 19, + "angle": 0, + "type": "image_body" + } + ], + "index": 19 + }, + { + "type": "image", + "bbox": [ + 414, + 266, + 456, + 306 + ], + "blocks": [ + { + "bbox": [ + 414, + 266, + 456, + 306 + ], + "lines": [ + { + "bbox": [ + 414, + 266, + 456, + 306 + ], + "spans": [ + { + "bbox": [ + 414, + 266, + 456, + 306 + ], + "type": "image", + "image_path": "997686e00c4a7784e1cdeb12ce40b00a953f11cf87dcc97cb21ebaf6cf1824d0.jpg" + } + ] + } + ], + "index": 20, + "angle": 0, + "type": "image_body" + } + ], + "index": 20 + }, + { + "type": "image", + "bbox": [ + 459, + 266, + 500, + 306 + ], + "blocks": [ + { + "bbox": [ + 459, + 266, + 500, + 306 + ], + "lines": [ + { + "bbox": [ + 459, + 266, + 500, + 306 + ], + "spans": [ + { + "bbox": [ + 459, + 266, + 500, + 306 + ], + "type": "image", + "image_path": "31d3aaf85111125ebeaf052529692cc6c988e95a07f05d51e6f1c22d159fafbc.jpg" + } + ] + } + ], + "index": 21, + "angle": 0, + "type": "image_body" + } + ], + "index": 21 + }, + { + "type": "image", + "bbox": [ + 503, + 266, + 544, + 306 + ], + "blocks": [ + { + "bbox": [ + 503, + 266, + 544, + 306 + ], + "lines": [ + { + "bbox": [ + 503, + 266, + 544, + 306 + ], + "spans": [ + { + "bbox": [ + 503, + 266, + 544, + 306 + ], + "type": "image", + "image_path": "8fa3e6cd82c0e860c844c50297115a37476eb73992c9f004202559cbe3ccce5d.jpg" + } + ] + } + ], + "index": 22, + "angle": 0, + "type": "image_body" + } + ], + "index": 22 + }, + { + "bbox": [ + 313, + 365, + 554, + 413 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 365, + 554, + 413 + ], + "spans": [ + { + "bbox": [ + 313, + 365, + 554, + 413 + ], + "type": "text", + "content": "additional training. Although the training-based method, MIGC, achieves the best FID score of 9.82, our method achieves a similarly FID of 10.03. The slight reduction can be attributed to the optimization on attention maps [47]." + } + ] + } + ], + "index": 25 + }, + { + "bbox": [ + 313, + 413, + 554, + 544 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 413, + 554, + 544 + ], + "spans": [ + { + "bbox": [ + 313, + 413, + 554, + 544 + ], + "type": "text", + "content": "User Study. We also conduct a user study with 63 participants to assess various methods through subjective evaluation. We present users with ten sets of 8 images along the same input conditions (i.e., prompt and layout map), and ask them to rank the given images based on image quality (AR-Q), layout (AR-L), and occlusion accuracy (AR-O). Each pair of images are shown randomly, and we collect ranking scores from different participants. We then compute the average ranking (AR) for three aspects: quality, layout, and occlusion as shown in Tab. 1. The results show a clear preference for our method among the users." + } + ] + } + ], + "index": 26 + }, + { + "bbox": [ + 313, + 546, + 408, + 558 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 546, + 408, + 558 + ], + "spans": [ + { + "bbox": [ + 313, + 546, + 408, + 558 + ], + "type": "text", + "content": "4.4. Ablation Study" + } + ] + } + ], + "index": 27 + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "spans": [ + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "text", + "content": "Component Analysis. We ablate our VODiff by removing SDP and VOA separately, and replacing the smooth guidance with constant guidance (denoted as " + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "inline_equation", + "content": "\\mathrm{VOA + SDP}_c" + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "text", + "content": "). Tab. 2 and Fig. 5 show the quantitative and qualitative results. We have three conclusions. Using only SDP (Dynamic) results in the worst LA, OA accuracy and Clip-Score but produces better FID. This is because SDP only adds visual guidance to guide the denoising direction, which does not impose stronger spatial constraints, and thus does not significantly affect the denoising process. VOA significantly improves the layout and occlusion accuracy by optimizing the cross-attention map, but at the same time, causes the model to produce less diversified results, thus impacting" + } + ] + } + ], + "index": 28 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "text", + "content": "18385" + } + ] + } + ], + "index": 29 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 6 + }, + { + "para_blocks": [ + { + "type": "table", + "bbox": [ + 64, + 96, + 288, + 176 + ], + "blocks": [ + { + "bbox": [ + 55, + 71, + 295, + 93 + ], + "lines": [ + { + "bbox": [ + 55, + 71, + 295, + 93 + ], + "spans": [ + { + "bbox": [ + 55, + 71, + 295, + 93 + ], + "type": "text", + "content": "Table 3. Ablation study on the VOA loss. " + }, + { + "bbox": [ + 55, + 71, + 295, + 93 + ], + "type": "inline_equation", + "content": "{\\mathcal{L}}_{\\mathrm{v}},{\\mathcal{L}}_{\\mathrm{o}}" + }, + { + "bbox": [ + 55, + 71, + 295, + 93 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 55, + 71, + 295, + 93 + ], + "type": "inline_equation", + "content": "{\\mathcal{L}}_{\\mathrm{b}}" + }, + { + "bbox": [ + 55, + 71, + 295, + 93 + ], + "type": "text", + "content": " represent the loss on visible, occluded and background regions, respectively." + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "table_caption" + }, + { + "bbox": [ + 64, + 96, + 288, + 176 + ], + "lines": [ + { + "bbox": [ + 64, + 96, + 288, + 176 + ], + "spans": [ + { + "bbox": [ + 64, + 96, + 288, + 176 + ], + "type": "table", + "html": "
LvLoLbFID ↓Clip-Score ↑LA ↑OA ↑
✓11.4128.0542.1360.50
✓✓10.2329.2546.4261.50
✓✓14.6528.4244.4777.50
✓✓11.6828.1029.1458.00
✓✓✓10.0329.7355.1182.50
", + "image_path": "dd20bf8d00fb9698e3590f4525d03862108af5f8676c16e5fb68b4693fef7b24.jpg" + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "table_body" + } + ], + "index": 1 + }, + { + "type": "image", + "bbox": [ + 67, + 195, + 115, + 240 + ], + "blocks": [ + { + "bbox": [ + 69, + 185, + 112, + 194 + ], + "lines": [ + { + "bbox": [ + 69, + 185, + 112, + 194 + ], + "spans": [ + { + "bbox": [ + 69, + 185, + 112, + 194 + ], + "type": "text", + "content": "Layout+Order" + } + ] + } + ], + "index": 2, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 67, + 195, + 115, + 240 + ], + "lines": [ + { + "bbox": [ + 67, + 195, + 115, + 240 + ], + "spans": [ + { + "bbox": [ + 67, + 195, + 115, + 240 + ], + "type": "image", + "image_path": "8536f1f2a099ef466ac33c162f3050e267cd2dec4c9d3031f865bd85476b664e.jpg" + } + ] + } + ], + "index": 4, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 67, + 245, + 113, + 277 + ], + "lines": [ + { + "bbox": [ + 67, + 245, + 113, + 277 + ], + "spans": [ + { + "bbox": [ + 67, + 245, + 113, + 277 + ], + "type": "text", + "content": "\"A man sitting in front of a table, a laptop and a cup on the table\"" + } + ] + } + ], + "index": 5, + "angle": 0, + "type": "image_caption" + } + ], + "index": 4 + }, + { + "type": "image", + "bbox": [ + 120, + 195, + 287, + 277 + ], + "blocks": [ + { + "bbox": [ + 120, + 185, + 292, + 194 + ], + "lines": [ + { + "bbox": [ + 120, + 185, + 292, + 194 + ], + "spans": [ + { + "bbox": [ + 120, + 185, + 292, + 194 + ], + "type": "text", + "content": "GLIGEN[30] ControlNet[79] MIGC[83] InstanceDiff.[65]" + } + ] + } + ], + "index": 3, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 120, + 195, + 287, + 277 + ], + "lines": [ + { + "bbox": [ + 120, + 195, + 287, + 277 + ], + "spans": [ + { + "bbox": [ + 120, + 195, + 287, + 277 + ], + "type": "image", + "image_path": "04ae46556d5dac17117f5acbc6f8fe93b109eacf9f266bca9425045c7ded64d0.jpg" + } + ] + } + ], + "index": 6, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 55, + 286, + 295, + 309 + ], + "lines": [ + { + "bbox": [ + 55, + 286, + 295, + 309 + ], + "spans": [ + { + "bbox": [ + 55, + 286, + 295, + 309 + ], + "type": "text", + "content": "Figure 6. Enhancement of existing methods using VODiff. Row 1 and 2 show the vanilla and the enhanced results, respectively." + } + ] + } + ], + "index": 7, + "angle": 0, + "type": "image_caption" + } + ], + "index": 6 + }, + { + "bbox": [ + 54, + 327, + 295, + 399 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 327, + 295, + 399 + ], + "spans": [ + { + "bbox": [ + 54, + 327, + 295, + 399 + ], + "type": "text", + "content": "the FID metrics. ⑧ Although constant visual guidance can improve FID, CLIP-Score, and LA, when combined with VOA, it confuses the order of objects in overlapping regions during the denoising process, interfering with attention optimization and worsening the OA. Finally, combining all the proposed components helps achieve the best results." + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 55, + 400, + 296, + 555 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 400, + 296, + 555 + ], + "spans": [ + { + "bbox": [ + 55, + 400, + 296, + 555 + ], + "type": "text", + "content": "Visibility-Order-Aware (VOA) loss. We verify the efficacy of different loss items in the VOA by removing them separately. Tab. 3 shows the quantitative results, and Supp. Fig. 3 shows the visual results. We can see that " + }, + { + "bbox": [ + 55, + 400, + 296, + 555 + ], + "type": "inline_equation", + "content": "\\mathbf{1} \\mathcal{L}_v" + }, + { + "bbox": [ + 55, + 400, + 296, + 555 + ], + "type": "text", + "content": " is indispensable for layout and occlusion constraints (4th row), but with " + }, + { + "bbox": [ + 55, + 400, + 296, + 555 + ], + "type": "inline_equation", + "content": "\\mathcal{L}_v" + }, + { + "bbox": [ + 55, + 400, + 296, + 555 + ], + "type": "text", + "content": " alone is also insufficient (1st row); " + }, + { + "bbox": [ + 55, + 400, + 296, + 555 + ], + "type": "inline_equation", + "content": "\\mathbf{2} \\mathcal{L}_v + \\mathcal{L}_b" + }, + { + "bbox": [ + 55, + 400, + 296, + 555 + ], + "type": "text", + "content": " produces inferior OA results as the overlapping regions lack constraints; " + }, + { + "bbox": [ + 55, + 400, + 296, + 555 + ], + "type": "inline_equation", + "content": "\\mathbf{3} \\mathcal{L}_v + \\mathcal{L}_o" + }, + { + "bbox": [ + 55, + 400, + 296, + 555 + ], + "type": "text", + "content": " has the remarkable OA results, but the worst FID because this combination only considers each object by itself but ignores background constraints (e.g., misaligned \"man\" and \"dog\" in the 1st row, 5-th column of Fig. 3 in Supp.). Finally, combining all items helps achieve the best performance on all metrics." + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 55, + 563, + 175, + 577 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 563, + 175, + 577 + ], + "spans": [ + { + "bbox": [ + 55, + 563, + 175, + 577 + ], + "type": "text", + "content": "4.5. Diverse Applications" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 55, + 581, + 295, + 700 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 581, + 295, + 700 + ], + "spans": [ + { + "bbox": [ + 55, + 581, + 295, + 700 + ], + "type": "text", + "content": "Our VODiff is training-free and can be easily integrated into many existing layout-to-image methods [30, 65, 79, 83], significantly enhancing the accuracy of occlusion relationship generation in current pretrained models. As shown in Fig. 6, the result from each method without (or with) using our approach is presented in the first (or second) row. Our VODiff effectively assists these training-based methods in addressing issues such as object disappearance (columns 2-4), as well as incorrect positional (column 5) and occlusion (columns 2 and 5) relationships." + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 67, + 701, + 295, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 67, + 701, + 295, + 713 + ], + "spans": [ + { + "bbox": [ + 67, + 701, + 295, + 713 + ], + "type": "text", + "content": "Our VODiff supports customized generation by integrat-" + } + ] + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 317, + 75, + 550, + 129 + ], + "blocks": [ + { + "bbox": [ + 317, + 75, + 550, + 129 + ], + "lines": [ + { + "bbox": [ + 317, + 75, + 550, + 129 + ], + "spans": [ + { + "bbox": [ + 317, + 75, + 550, + 129 + ], + "type": "image", + "image_path": "9de675883694e7e7c6e6fd7d0049d94fa2cf0b4b7498f0e0a221bb5a6fae2665.jpg" + } + ] + } + ], + "index": 13, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 372, + 130, + 498, + 138 + ], + "lines": [ + { + "bbox": [ + 372, + 130, + 498, + 138 + ], + "spans": [ + { + "bbox": [ + 372, + 130, + 498, + 138 + ], + "type": "text", + "content": "(a) Object insertion with accurate visibility orders" + } + ] + } + ], + "index": 14, + "angle": 0, + "type": "image_caption" + } + ], + "index": 13 + }, + { + "type": "image", + "bbox": [ + 317, + 139, + 550, + 194 + ], + "blocks": [ + { + "bbox": [ + 317, + 139, + 550, + 194 + ], + "lines": [ + { + "bbox": [ + 317, + 139, + 550, + 194 + ], + "spans": [ + { + "bbox": [ + 317, + 139, + 550, + 194 + ], + "type": "image", + "image_path": "15a480cd80dce19f2db67f37c02eb28b9b5290e460103822fc6e5957453cf8e3.jpg" + } + ] + } + ], + "index": 15, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 365, + 195, + 504, + 204 + ], + "lines": [ + { + "bbox": [ + 365, + 195, + 504, + 204 + ], + "spans": [ + { + "bbox": [ + 365, + 195, + 504, + 204 + ], + "type": "text", + "content": "(b) Exemplar based image editing with visibility orders" + } + ] + } + ], + "index": 16, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 313, + 209, + 553, + 232 + ], + "lines": [ + { + "bbox": [ + 313, + 209, + 553, + 232 + ], + "spans": [ + { + "bbox": [ + 313, + 209, + 553, + 232 + ], + "type": "text", + "content": "Figure 7. Customized generation with VODiff in object insertion (row 1) and exemplar-based editing (row 2)." + } + ] + } + ], + "index": 17, + "angle": 0, + "type": "image_caption" + } + ], + "index": 15 + }, + { + "bbox": [ + 313, + 238, + 555, + 358 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 238, + 555, + 358 + ], + "spans": [ + { + "bbox": [ + 313, + 238, + 555, + 358 + ], + "type": "text", + "content": "ing it with existing techniques, e.g., DreamBooth [56]. In Fig. 7(a), we first demonstrate the insertion of newly generated objects into an existing image via Inversion [59], where the inserted objects adhere to the preset visibility order relative to the original objects in the image, while preserving the background unchanged. Fig. 7(b) shows the integration of our method with DreamBooth [56], enabling the generated objects to not only maintain the identity of the input objects but also ensure that the positions and visibility order among multiple objects adhere to the preset arrangement." + } + ] + } + ], + "index": 18 + }, + { + "type": "image", + "bbox": [ + 328, + 366, + 541, + 443 + ], + "blocks": [ + { + "bbox": [ + 328, + 366, + 541, + 443 + ], + "lines": [ + { + "bbox": [ + 328, + 366, + 541, + 443 + ], + "spans": [ + { + "bbox": [ + 328, + 366, + 541, + 443 + ], + "type": "image", + "image_path": "92bcd44445f22b7da1ba9d6868606ec75fa9a0ddec913bdea472536451d1b417.jpg" + } + ] + } + ], + "index": 19, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 333, + 441, + 537, + 467 + ], + "lines": [ + { + "bbox": [ + 333, + 441, + 537, + 467 + ], + "spans": [ + { + "bbox": [ + 333, + 441, + 537, + 467 + ], + "type": "text", + "content": "\"A bedroom with a cabinet behind a plant, and there is also a bed, a flower on the right and a boy in front of a lamp who is feeding a dog, and a glass bottle is in front of him.\"" + } + ] + } + ], + "index": 20, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 313, + 475, + 553, + 498 + ], + "lines": [ + { + "bbox": [ + 313, + 475, + 553, + 498 + ], + "spans": [ + { + "bbox": [ + 313, + 475, + 553, + 498 + ], + "type": "text", + "content": "Figure 8. A failure case of VODiff. From left to right are: inputs, results of InstanceDiff. [65] and Our method, respectively." + } + ] + } + ], + "index": 21, + "angle": 0, + "type": "image_caption" + } + ], + "index": 19 + }, + { + "bbox": [ + 313, + 507, + 388, + 520 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 507, + 388, + 520 + ], + "spans": [ + { + "bbox": [ + 313, + 507, + 388, + 520 + ], + "type": "text", + "content": "5. Conclusion" + } + ] + } + ], + "index": 22 + }, + { + "bbox": [ + 313, + 521, + 554, + 653 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 521, + 554, + 653 + ], + "spans": [ + { + "bbox": [ + 313, + 521, + 554, + 653 + ], + "type": "text", + "content": "In this paper, we have introduced VODiff to address the challenges of generating complex scenes with object occlusions. We first propose a Sequential Denoising Process that performs object generation sequentially from the bottom to the top layer based on the visibility order of objects. We then propose a Visibility-Order-Aware loss to constrain object positions and occlusion relationships by transforming the layout and visibility order into cross-attention map optimization. The results on our curated VOBench show superior performances of our method in controlling object spatial position and occlusion relationships." + } + ] + } + ], + "index": 23 + }, + { + "bbox": [ + 313, + 654, + 554, + 714 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 654, + 554, + 714 + ], + "spans": [ + { + "bbox": [ + 313, + 654, + 554, + 714 + ], + "type": "text", + "content": "VODiff does has limitations. For example, as shown in Fig. 8, like existing method [65], our method may struggle to handle scenarios where the number of objects is exceptionally high due to the challenge of balancing the image quality with precise spatial and occlusion relationships." + } + ] + } + ], + "index": 24 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "text", + "content": "18386" + } + ] + } + ], + "index": 25 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 7 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 72, + 158, + 85 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 72, + 158, + 85 + ], + "spans": [ + { + "bbox": [ + 56, + 72, + 158, + 85 + ], + "type": "text", + "content": "Acknowledgements" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 91, + 296, + 140 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 91, + 296, + 140 + ], + "spans": [ + { + "bbox": [ + 55, + 91, + 296, + 140 + ], + "type": "text", + "content": "This work is supported in part by Ant Group and General Program of National Natural Science Foundation of China (NSFC, No. 6207071897) and a GRF from the Research Grants Council of Hong Kong (Ref.: 11205620)." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 148, + 115, + 161 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 148, + 115, + 161 + ], + "spans": [ + { + "bbox": [ + 56, + 148, + 115, + 161 + ], + "type": "text", + "content": "References" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 57, + 168, + 296, + 713 + ], + "type": "list", + "angle": 0, + "index": 16, + "blocks": [ + { + "bbox": [ + 61, + 168, + 296, + 223 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 61, + 168, + 296, + 223 + ], + "spans": [ + { + "bbox": [ + 61, + 168, + 296, + 223 + ], + "type": "text", + "content": "[1] Omri Avrahami, Thomas Hayes, Oran Gafni, Sonal Gupta, Yaniv Taigman, Devi Parikh, Dani Lischinski, Ohad Fried, and Xi Yin. Spatext: Spatio-textual representation for controllable image generation. In CVPR, pages 18370-18380, 2023. 3" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 61, + 224, + 296, + 277 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 61, + 224, + 296, + 277 + ], + "spans": [ + { + "bbox": [ + 61, + 224, + 296, + 277 + ], + "type": "text", + "content": "[2] Yogesh Balaji, Seungjun Nah, Xun Huang, Arash Vahdat, Jiaming Song, Qinsheng Zhang, Karsten Kreis, Miika Aittala, Timo Aila, Samuli Laine, et al. ediff-i: Text-to-image diffusion models with an ensemble of expert denoisers. arXiv:2211.01324, 2022. 3.5" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 61, + 280, + 295, + 312 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 61, + 280, + 295, + 312 + ], + "spans": [ + { + "bbox": [ + 61, + 280, + 295, + 312 + ], + "type": "text", + "content": "[3] Omer Bar-Tal, Lior Yariv, Yaron Lipman, and Tali Dekel. Multidiffusion: fusing diffusion paths for controlled image generation. In ICML, pages 1737-1752, 2023. 2, 3, 7, 5" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 62, + 314, + 294, + 346 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 314, + 294, + 346 + ], + "spans": [ + { + "bbox": [ + 62, + 314, + 294, + 346 + ], + "type": "text", + "content": "[4] Tim Brooks, Aleksander Holynski, and Alexei A Efros. Instructpix2pix: Learning to follow image editing instructions. In CVPR, pages 18392-18402, 2023. 2" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 62, + 347, + 295, + 389 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 347, + 295, + 389 + ], + "spans": [ + { + "bbox": [ + 62, + 347, + 295, + 389 + ], + "type": "text", + "content": "[5] Hila Chefer, Yuval Alaluf, Yael Vinker, Lior Wolf, and Daniel Cohen-Or. Attend-and-excite: Attention-based semantic guidance for text-to-image diffusion models. ACM TOG, 42(4):1-10, 2023. 3, 1" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 62, + 391, + 295, + 434 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 391, + 295, + 434 + ], + "spans": [ + { + "bbox": [ + 62, + 391, + 295, + 434 + ], + "type": "text", + "content": "[6] Hong Chen, Yipeng Zhang, Xin Wang, Xuguang Duan, Yuwei Zhou, and Wenwu Zhu. Disenbooth: Disentangled parameter-efficient tuning for subject-driven text-to-image generation. arXiv:2305.03374, 3, 2023. 1" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 62, + 436, + 294, + 479 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 436, + 294, + 479 + ], + "spans": [ + { + "bbox": [ + 62, + 436, + 294, + 479 + ], + "type": "text", + "content": "[7] Kai Chen, Enze Xie, Zhe Chen, Yibo Wang, HONG Lanqing, Zhenguo Li, and Dit-Yan Yeung. Geodiffusion: Text-prompted geometric control for object detection data generation. In ICLR, 2024. 3" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 62, + 480, + 294, + 513 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 480, + 294, + 513 + ], + "spans": [ + { + "bbox": [ + 62, + 480, + 294, + 513 + ], + "type": "text", + "content": "[8] Minghao Chen, Iro Laina, and Andrea Vedaldi. Training-free layout control with cross-attention guidance. In WACV, pages 5343-5353, 2024. 3, 2" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 62, + 514, + 294, + 555 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 514, + 294, + 555 + ], + "spans": [ + { + "bbox": [ + 62, + 514, + 294, + 555 + ], + "type": "text", + "content": "[9] Xi Chen, Lianghua Huang, Yu Liu, Yujun Shen, Deli Zhao, and Hengshuang Zhao. Anydoor: Zero-shot object-level image customization. In CVPR, pages 6593-6602, 2024. 3, 1, 6" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 57, + 558, + 295, + 601 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 558, + 295, + 601 + ], + "spans": [ + { + "bbox": [ + 57, + 558, + 295, + 601 + ], + "type": "text", + "content": "[10] Jiaxin Cheng, Xiao Liang, Xingjian Shi, Tong He, Tianjun Xiao, and Mu Li. Layoutdiffuse: Adapting foundational diffusion models for layout-to-image generation. arXiv preprint arXiv:2302.08908, 2023. 3" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 57, + 602, + 294, + 635 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 602, + 294, + 635 + ], + "spans": [ + { + "bbox": [ + 57, + 602, + 294, + 635 + ], + "type": "text", + "content": "[11] Patrick Esser, Robin Rombach, and Bjorn Ommer. Taming transformers for high-resolution image synthesis. In CVPR, pages 12873-12883, 2021. 3" + } + ] + } + ], + "index": 13 + }, + { + "bbox": [ + 57, + 636, + 294, + 689 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 636, + 294, + 689 + ], + "spans": [ + { + "bbox": [ + 57, + 636, + 294, + 689 + ], + "type": "text", + "content": "[12] Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. In ICML, 2024. 3" + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 57, + 691, + 294, + 713 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 691, + 294, + 713 + ], + "spans": [ + { + "bbox": [ + 57, + 691, + 294, + 713 + ], + "type": "text", + "content": "[13] Weixi Feng, Xuehai He, Tsu-Jui Fu, Varun Jampani, Arjun Akula, Pradyumna Narayana, Sugato Basu, Xin Eric" + } + ] + } + ], + "index": 15 + } + ], + "sub_type": "ref_text" + }, + { + "bbox": [ + 316, + 73, + 553, + 712 + ], + "type": "list", + "angle": 0, + "index": 32, + "blocks": [ + { + "bbox": [ + 335, + 73, + 553, + 106 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 335, + 73, + 553, + 106 + ], + "spans": [ + { + "bbox": [ + 335, + 73, + 553, + 106 + ], + "type": "text", + "content": "Wang, and William Yang Wang. Training-free structured diffusion guidance for compositional text-to-image synthesis. arXiv:2212.05032, 2022. 3" + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 316, + 108, + 553, + 152 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 108, + 553, + 152 + ], + "spans": [ + { + "bbox": [ + 316, + 108, + 553, + 152 + ], + "type": "text", + "content": "[14] Rinon Gal, Yuval Alaluf, Yuval Atzmon, Or Patashnik, Amit H Bermano, Gal Chechik, and Daniel Cohen-Or. An image is worth one word: Personalizing text-to-image generation using textual inversion. arXiv:2208.01618, 2022. 1" + } + ] + } + ], + "index": 18 + }, + { + "bbox": [ + 316, + 153, + 553, + 197 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 153, + 553, + 197 + ], + "spans": [ + { + "bbox": [ + 316, + 153, + 553, + 197 + ], + "type": "text", + "content": "[15] Biao Gong, Siteng Huang, Yutong Feng, Shiwei Zhang, Yuyuan Li, and Yu Liu. Check locate rectify: A training-free layout calibration system for text-to-image generation. In CVPR, pages 6624-6634, 2024. 3" + } + ] + } + ], + "index": 19 + }, + { + "bbox": [ + 316, + 199, + 553, + 243 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 199, + 553, + 243 + ], + "spans": [ + { + "bbox": [ + 316, + 199, + 553, + 243 + ], + "type": "text", + "content": "[16] Shuyang Gu, Dong Chen, Jianmin Bao, Fang Wen, Bo Zhang, Dongdong Chen, Lu Yuan, and Baining Guo. Vector quantized diffusion model for text-to-image synthesis. In CVPR, pages 10696-10706, 2022. 3" + } + ] + } + ], + "index": 20 + }, + { + "bbox": [ + 316, + 245, + 553, + 277 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 245, + 553, + 277 + ], + "spans": [ + { + "bbox": [ + 316, + 245, + 553, + 277 + ], + "type": "text", + "content": "[17] Amir Hertz, Ron Mokady, Jay Tenenbaum, Kfir Aberman, Yael Pritch, and Daniel Cohen-or. Prompt-to-prompt image editing with cross-attention control. In ICLR, 2023. 5" + } + ] + } + ], + "index": 21 + }, + { + "bbox": [ + 316, + 280, + 553, + 323 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 280, + 553, + 323 + ], + "spans": [ + { + "bbox": [ + 316, + 280, + 553, + 323 + ], + "type": "text", + "content": "[18] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In NIPS, 2017. 6" + } + ] + } + ], + "index": 22 + }, + { + "bbox": [ + 316, + 326, + 553, + 347 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 326, + 553, + 347 + ], + "spans": [ + { + "bbox": [ + 316, + 326, + 553, + 347 + ], + "type": "text", + "content": "[19] Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. In NIPS Workshop, 2021. 5" + } + ] + } + ], + "index": 23 + }, + { + "bbox": [ + 316, + 350, + 553, + 371 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 350, + 553, + 371 + ], + "spans": [ + { + "bbox": [ + 316, + 350, + 553, + 371 + ], + "type": "text", + "content": "[20] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In NIPS, 2020. 2, 3" + } + ] + } + ], + "index": 24 + }, + { + "bbox": [ + 316, + 373, + 553, + 416 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 373, + 553, + 416 + ], + "spans": [ + { + "bbox": [ + 316, + 373, + 553, + 416 + ], + "type": "text", + "content": "[21] Jonathan Ho, Chitwan Sahara, William Chan, David J Fleet, Mohammad Norouzi, and Tim Salimans. Cascaded diffusion models for high fidelity image generation. JMLR, 23(47):1-33, 2022. 3" + } + ] + } + ], + "index": 25 + }, + { + "bbox": [ + 316, + 418, + 553, + 472 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 418, + 553, + 472 + ], + "spans": [ + { + "bbox": [ + 316, + 418, + 553, + 472 + ], + "type": "text", + "content": "[22] Hexiang Hu, Kelvin CK Chan, Yu-Chuan Su, Wenhu Chen, Yandong Li, Kihyuk Sohn, Yang Zhao, Xue Ben, Boqing Gong, William Cohen, et al. Instruct-imagen: Image generation with multi-modal instruction. In CVPR, pages 4754-4763, 2024. 3" + } + ] + } + ], + "index": 26 + }, + { + "bbox": [ + 316, + 475, + 553, + 529 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 475, + 553, + 529 + ], + "spans": [ + { + "bbox": [ + 316, + 475, + 553, + 529 + ], + "type": "text", + "content": "[23] Chengyou Jia, Minnan Luo, Zhuohang Dang, Guang Dai, Xiaojun Chang, Mengmeng Wang, and Jingdong Wang. Ssmg: Spatial-semantic map guided diffusion model for free-form layout-to-image generation. In AAAI, pages 2480–2488, 2024. 3" + } + ] + } + ], + "index": 27 + }, + { + "bbox": [ + 316, + 533, + 553, + 575 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 533, + 553, + 575 + ], + "spans": [ + { + "bbox": [ + 316, + 533, + 553, + 575 + ], + "type": "text", + "content": "[24] Yuming Jiang, Nanxuan Zhao, Qing Liu, Krishna Kumar Singh, Shuai Yang, Chen Change Loy, and Ziwei Liu. Groupdiff: Diffusion-based group portrait editing. arXiv:2409.14379, 2024. 1" + } + ] + } + ], + "index": 28 + }, + { + "bbox": [ + 316, + 578, + 553, + 621 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 578, + 553, + 621 + ], + "spans": [ + { + "bbox": [ + 316, + 578, + 553, + 621 + ], + "type": "text", + "content": "[25] Zeyinzi Jiang, Chaojie Mao, Yulin Pan, Zhen Han, and Jingfeng Zhang. Scedit: Efficient and controllable image diffusion generation via skip connection editing. In CVPR, pages 8995–9004, 2024. 3" + } + ] + } + ], + "index": 29 + }, + { + "bbox": [ + 316, + 624, + 553, + 666 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 624, + 553, + 666 + ], + "spans": [ + { + "bbox": [ + 316, + 624, + 553, + 666 + ], + "type": "text", + "content": "[26] Sungnyun Kim, Junsoo Lee, Kibeom Hong, Daesik Kim, and Namhyuk Ahn. Diffblender: Scalable and composable multimodal text-to-image diffusion models. arXiv:2305.15194, 2023. 3" + } + ] + } + ], + "index": 30 + }, + { + "bbox": [ + 316, + 670, + 553, + 712 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 670, + 553, + 712 + ], + "spans": [ + { + "bbox": [ + 316, + 670, + 553, + 712 + ], + "type": "text", + "content": "[27] Yunji Kim, Jiyoung Lee, Jin-Hwa Kim, Jung-Woo Ha, and Jun-Yan Zhu. Dense text-to-image generation with attention modulation. In ICCV, pages 7701-7711, 2023. 1, 2, 3, 4, 7, 5" + } + ] + } + ], + "index": 31 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 294, + 748, + 317, + 757 + ], + "type": "text", + "content": "18387" + } + ] + } + ], + "index": 33 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 8 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 72, + 294, + 713 + ], + "type": "list", + "angle": 0, + "index": 14, + "blocks": [ + { + "bbox": [ + 56, + 72, + 294, + 116 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 72, + 294, + 116 + ], + "spans": [ + { + "bbox": [ + 56, + 72, + 294, + 116 + ], + "type": "text", + "content": "[28] Leheng Li, Weichao Qiu, Xu Yan, Jing He, Kaiqiang Zhou, Yingjie Cai, Qing Lian, Bingbing Liu, and Ying-Cong Chen. Omnibooth: Learning latent control for image synthesis with multi-modal instruction. arXiv:2410.04932, 2024. 1" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 118, + 294, + 140 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 118, + 294, + 140 + ], + "spans": [ + { + "bbox": [ + 56, + 118, + 294, + 140 + ], + "type": "text", + "content": "[29] Tianle Li, Max Ku, Cong Wei, and Wenhu Chen. Dreamedit: Subject-driven image editing. arXiv:2306.12624, 2023. 3, 1" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 141, + 294, + 185 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 141, + 294, + 185 + ], + "spans": [ + { + "bbox": [ + 56, + 141, + 294, + 185 + ], + "type": "text", + "content": "[30] Yuheng Li, Haotian Liu, Qingyang Wu, Fangzhou Mu, Jianwei Yang, Jianfeng Gao, Chunyuan Li, and Yong Jae Lee. Gligen: Open-set grounded text-to-image generation. In CVPR, pages 22511-22521, 2023. 1, 2, 3, 8, 5, 6, 9" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 56, + 186, + 294, + 239 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 186, + 294, + 239 + ], + "spans": [ + { + "bbox": [ + 56, + 186, + 294, + 239 + ], + "type": "text", + "content": "[31] Yaqi Li, Han Fang, Zerun Feng, Kaijing Ma, Chao Ban, Xianghao Zang, LanXiang Zhou, Zhongjiang He, Jingyan Chen, Jiani Hu, et al. Goal: Grounded text-to-image synthesis with joint layout alignment tuning. In ACM MM, pages 7055-7064, 2024. 3" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 56, + 242, + 294, + 275 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 242, + 294, + 275 + ], + "spans": [ + { + "bbox": [ + 56, + 242, + 294, + 275 + ], + "type": "text", + "content": "[32] Chang Liu, Xiangtai Li, and Henghui Ding. Referring image editing: Object-level image editing via referring expressions. In CVPR, pages 13128-13138, 2024. 3, 1" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 56, + 277, + 294, + 319 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 277, + 294, + 319 + ], + "spans": [ + { + "bbox": [ + 56, + 277, + 294, + 319 + ], + "type": "text", + "content": "[33] Nan Liu, Shuang Li, Yilun Du, Antonio Torralba, and Joshua B Tenenbaum. Compositional visual generation with composable diffusion models. In ECCV, pages 423-439, 2022. 3" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 56, + 321, + 294, + 375 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 321, + 294, + 375 + ], + "spans": [ + { + "bbox": [ + 56, + 321, + 294, + 375 + ], + "type": "text", + "content": "[34] Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. arXiv:2303.05499, 2023.6.1" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 56, + 377, + 294, + 421 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 377, + 294, + 421 + ], + "spans": [ + { + "bbox": [ + 56, + 377, + 294, + 421 + ], + "type": "text", + "content": "[35] Xiaoyu Liu, Yuxiang Wei, Ming Liu, Xianhui Lin, Peiran Ren, Xuansong Xie, and Wangmeng Zuo. Smartcontrol: Enhancing controlnet for handling rough visual conditions. arXiv:2404.06451, 2024. 3, 7" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 56, + 422, + 294, + 464 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 422, + 294, + 464 + ], + "spans": [ + { + "bbox": [ + 56, + 422, + 294, + 464 + ], + "type": "text", + "content": "[36] Yuhao Liu, Zhanghan Ke, Fang Liu, Nanxuan Zhao, and Rynson WH Lau. Diff-plugin: Revitalizing details for diffusion-based low-level tasks. In CVPR, pages 4197-4208, 2024. 3" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 56, + 468, + 294, + 510 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 468, + 294, + 510 + ], + "spans": [ + { + "bbox": [ + 56, + 468, + 294, + 510 + ], + "type": "text", + "content": "[37] Grace Luo, Trevor Darrell, Oliver Wang, Dan B Goldman, and Aleksander Holynski. Readout guidance: Learning control from diffusion features. In CVPR, pages 8217-8227, 2024. 3" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 56, + 512, + 294, + 566 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 512, + 294, + 566 + ], + "spans": [ + { + "bbox": [ + 56, + 512, + 294, + 566 + ], + "type": "text", + "content": "[38] Kangfu Mei, Mauricio Delbracio, Hossein Talebi, Zhengzhong Tu, Vishal M Patel, and Peyman Milanfar. Codi: Conditional diffusion distillation for higher-fidelity and faster image generation. In CVPR, pages 9048-9058, 2024. 3" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 56, + 568, + 294, + 622 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 568, + 294, + 622 + ], + "spans": [ + { + "bbox": [ + 56, + 568, + 294, + 622 + ], + "type": "text", + "content": "[39] Sicheng Mo, Fangzhou Mu, Kuan Heng Lin, Yanli Liu, Bochen Guan, Yin Li, and Bolei Zhou. Freecontrol: Training-free spatial control of any text-to-image diffusion model with any condition. In CVPR, pages 7465-7475, 2024. 3, 4, 6, 7, 5" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 56, + 624, + 294, + 667 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 624, + 294, + 667 + ], + "spans": [ + { + "bbox": [ + 56, + 624, + 294, + 667 + ], + "type": "text", + "content": "[40] Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In AAAI, pages 4296-4304, 2024. 3" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 56, + 670, + 294, + 713 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 670, + 294, + 713 + ], + "spans": [ + { + "bbox": [ + 56, + 670, + 294, + 713 + ], + "type": "text", + "content": "[41] Nithin Gopalakrishnan Nair, Jeya Maria Jose Valanarasu, and Vishal M Patel. Maxfusion: Plug&play multi-modal generation in text-to-image diffusion models. In ECCV, pages 93-110, 2025. 3" + } + ] + } + ], + "index": 13 + } + ], + "sub_type": "ref_text" + }, + { + "bbox": [ + 315, + 72, + 553, + 713 + ], + "type": "list", + "angle": 0, + "index": 30, + "blocks": [ + { + "bbox": [ + 315, + 72, + 553, + 115 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 72, + 553, + 115 + ], + "spans": [ + { + "bbox": [ + 315, + 72, + 553, + 115 + ], + "type": "text", + "content": "[42] Martin E Newell, RG Newell, and Tom L Sancha. A solution to the hidden surface problem. In Seminal graphics: pioneering efforts that shaped the field, pages 27-32. 1998. 4" + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 316, + 118, + 553, + 172 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 118, + 553, + 172 + ], + "spans": [ + { + "bbox": [ + 316, + 118, + 553, + 172 + ], + "type": "text", + "content": "[43] Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. Glide: Towards photorealistic image generation and editing with text-guided diffusion models. arXiv:2112.10741, 2021. 3" + } + ] + } + ], + "index": 16 + }, + { + "bbox": [ + 316, + 175, + 539, + 185 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 175, + 539, + 185 + ], + "spans": [ + { + "bbox": [ + 316, + 175, + 539, + 185 + ], + "type": "text", + "content": "[44] NovelAI. Novelai: Ai generated stories and art. 2021. 2" + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 317, + 186, + 553, + 228 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 186, + 553, + 228 + ], + "spans": [ + { + "bbox": [ + 317, + 186, + 553, + 228 + ], + "type": "text", + "content": "[45] Lingzhi Pan, Tong Zhang, Bingyuan Chen, Qi Zhou, Wei Ke, Sabine Süssstrunk, and Mathieu Salzmann. Coherent and multi-modality image inpainting via latent space optimization. arXiv:2407.08019, 2024. 1" + } + ] + } + ], + "index": 18 + }, + { + "bbox": [ + 317, + 231, + 553, + 274 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 231, + 553, + 274 + ], + "spans": [ + { + "bbox": [ + 317, + 231, + 553, + 274 + ], + "type": "text", + "content": "[46] Yulin Pan, Chaojie Mao, Zeyinzi Jiang, Zhen Han, and Jingfeng Zhang. Locate, assign, refine: Taming customized image inpainting with text-subject guidance. arXiv:2403.19534, 2024. 1" + } + ] + } + ], + "index": 19 + }, + { + "bbox": [ + 317, + 276, + 553, + 308 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 276, + 553, + 308 + ], + "spans": [ + { + "bbox": [ + 317, + 276, + 553, + 308 + ], + "type": "text", + "content": "[47] Quynh Phung, Songwei Ge, and Jia-Bin Huang. Grounded text-to-image synthesis with attention refocusing. In CVPR, pages 7932-7942, 2024. 2, 3, 4, 6, 7, 5" + } + ] + } + ], + "index": 20 + }, + { + "bbox": [ + 317, + 310, + 553, + 363 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 310, + 553, + 363 + ], + "spans": [ + { + "bbox": [ + 317, + 310, + 553, + 363 + ], + "type": "text", + "content": "[48] Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv:2307.01952, 2023. 3, 5" + } + ] + } + ], + "index": 21 + }, + { + "bbox": [ + 316, + 365, + 553, + 398 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 365, + 553, + 398 + ], + "spans": [ + { + "bbox": [ + 316, + 365, + 553, + 398 + ], + "type": "text", + "content": "[49] Zipeng Qi, Guoxi Huang, Chenyang Liu, and Fei Ye. Layered rendering diffusion model for zero-shot guided image synthesis. In ECCV, 2024. 3, 1" + } + ] + } + ], + "index": 22 + }, + { + "bbox": [ + 317, + 399, + 553, + 453 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 399, + 553, + 453 + ], + "spans": [ + { + "bbox": [ + 317, + 399, + 553, + 453 + ], + "type": "text", + "content": "[50] Can Qin, Shu Zhang, Ning Yu, Yihao Feng, Xinyi Yang, Yingbo Zhou, Huan Wang, Juan Carlos Niebles, Caiming Xiong, Silvio Savarese, et al. Uniconcontrol: A unified diffusion model for controllable visual generation in the wild. In NIPS, 2023. 3" + } + ] + } + ], + "index": 23 + }, + { + "bbox": [ + 317, + 456, + 553, + 498 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 456, + 553, + 498 + ], + "spans": [ + { + "bbox": [ + 317, + 456, + 553, + 498 + ], + "type": "text", + "content": "[51] Leigang Qu, Shengqiong Wu, Hao Fei, Liqiang Nie, and TatSeng Chua. Layoutllm-t2i: Eliciting layout guidance from llm for text-to-image generation. In ACM MM, pages 643-654, 2023. 3" + } + ] + } + ], + "index": 24 + }, + { + "bbox": [ + 317, + 500, + 553, + 555 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 500, + 553, + 555 + ], + "spans": [ + { + "bbox": [ + 317, + 500, + 553, + 555 + ], + "type": "text", + "content": "[52] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In ICML, pages 8748-8763, 2021. 3, 4" + } + ] + } + ], + "index": 25 + }, + { + "bbox": [ + 317, + 556, + 553, + 609 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 556, + 553, + 609 + ], + "spans": [ + { + "bbox": [ + 317, + 556, + 553, + 609 + ], + "type": "text", + "content": "[53] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. JMLR, 21(140):1-67, 2020. 3" + } + ] + } + ], + "index": 26 + }, + { + "bbox": [ + 317, + 612, + 553, + 644 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 612, + 553, + 644 + ], + "spans": [ + { + "bbox": [ + 317, + 612, + 553, + 644 + ], + "type": "text", + "content": "[54] Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. arXiv:2204.06125, 2022. 2, 3" + } + ] + } + ], + "index": 27 + }, + { + "bbox": [ + 317, + 647, + 553, + 689 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 647, + 553, + 689 + ], + "spans": [ + { + "bbox": [ + 317, + 647, + 553, + 689 + ], + "type": "text", + "content": "[55] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In CVPR, pages 10684-10695, 2022. 1, 2, 3, 4, 5" + } + ] + } + ], + "index": 28 + }, + { + "bbox": [ + 316, + 691, + 553, + 713 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 691, + 553, + 713 + ], + "spans": [ + { + "bbox": [ + 316, + 691, + 553, + 713 + ], + "type": "text", + "content": "[56] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine" + } + ] + } + ], + "index": 29 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 294, + 749, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 294, + 749, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 294, + 749, + 317, + 757 + ], + "type": "text", + "content": "18388" + } + ] + } + ], + "index": 31 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 9 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 73, + 295, + 712 + ], + "type": "list", + "angle": 0, + "index": 14, + "blocks": [ + { + "bbox": [ + 76, + 73, + 294, + 95 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 76, + 73, + 294, + 95 + ], + "spans": [ + { + "bbox": [ + 76, + 73, + 294, + 95 + ], + "type": "text", + "content": "tuning text-to-image diffusion models for subject-driven generation. In CVPR, pages 22500-22510, 2023. 8, 1, 6" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 96, + 295, + 162 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 96, + 295, + 162 + ], + "spans": [ + { + "bbox": [ + 56, + 96, + 295, + 162 + ], + "type": "text", + "content": "[57] Chitwan Sahara, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding. In NIPS, pages 36479-36494, 2022. 2, 3" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 163, + 295, + 228 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 163, + 295, + 228 + ], + "spans": [ + { + "bbox": [ + 56, + 163, + 295, + 228 + ], + "type": "text", + "content": "[58] Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Wortsman, et al. Laion-5b: An open large-scale dataset for training next generation image-text models. In NIPS, pages 25278–25294, 2022. 3, 1" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 56, + 230, + 295, + 262 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 230, + 295, + 262 + ], + "spans": [ + { + "bbox": [ + 56, + 230, + 295, + 262 + ], + "type": "text", + "content": "[59] Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. In ICLR, 2021. 2, 3, 8, 1, 6" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 56, + 264, + 295, + 308 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 264, + 295, + 308 + ], + "spans": [ + { + "bbox": [ + 56, + 264, + 295, + 308 + ], + "type": "text", + "content": "[60] Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In ICLR, 2021. 2, 3" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 56, + 309, + 295, + 353 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 309, + 295, + 353 + ], + "spans": [ + { + "bbox": [ + 56, + 309, + 295, + 353 + ], + "type": "text", + "content": "[61] Yizhi Song, Zhifei Zhang, Zhe Lin, Scott Cohen, Brian Price, Jianming Zhang, Soo Ye Kim, and Daniel Aliaga. Object-stitch: Object compositing with diffusion model. In CVPR, pages 18310-18319, 2023. 3, 1" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 56, + 354, + 295, + 408 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 354, + 295, + 408 + ], + "spans": [ + { + "bbox": [ + 56, + 354, + 295, + 408 + ], + "type": "text", + "content": "[62] Yizhi Song, Zhifei Zhang, Zhe Lin, Scott Cohen, Brian Price, Jianming Zhang, Soo Ye Kim, He Zhang, Wei Xiong, and Daniel Aliaga. Imprint: Generative object compositing by learning identity-preserving representation. In CVPR, pages 8048-8058, 2024. 3, 1, 6" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 56, + 410, + 295, + 443 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 410, + 295, + 443 + ], + "spans": [ + { + "bbox": [ + 56, + 410, + 295, + 443 + ], + "type": "text", + "content": "[63] Yanan Sun, Yanchen Liu, Yinhao Tang, Wenjie Pei, and Kai Chen. Anycontrol: Create your artwork with versatile control on text-to-image generation, 2024. 3, 6, 7" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 56, + 445, + 295, + 488 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 445, + 295, + 488 + ], + "spans": [ + { + "bbox": [ + 56, + 445, + 295, + 488 + ], + "type": "text", + "content": "[64] Zhenhong Sun, Junyan Wang, Zhiyu Tan, Daoyi Dong, Hailan Ma, Hao Li, and Dong Gong. Eggen: Image generation with multi-entity prior learning through entity guidance. In ACM MM, pages 6637-6645, 2024. 3" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 56, + 490, + 295, + 533 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 490, + 295, + 533 + ], + "spans": [ + { + "bbox": [ + 56, + 490, + 295, + 533 + ], + "type": "text", + "content": "[65] Xudong Wang, Trevor Darrell, Sai Saketh Rambhatla, Rohit Girdhar, and Ishan Misra. Instancediffusion: Instance-level control for image generation. In CVPR, pages 6232-6242, 2024. 3, 6, 7, 8, 5, 9" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 56, + 535, + 295, + 578 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 535, + 295, + 578 + ], + "spans": [ + { + "bbox": [ + 56, + 535, + 295, + 578 + ], + "type": "text", + "content": "[66] Yuxiang Wei, Yabo Zhang, Zhilong Ji, Jinfeng Bai, Lei Zhang, and Wangmeng Zuo. Elite: Encoding visual concepts into textual embeddings for customized text-to-image generation. In ICCV, pages 15943-15953, 2023. 1" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 56, + 580, + 295, + 622 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 580, + 295, + 622 + ], + "spans": [ + { + "bbox": [ + 56, + 580, + 295, + 622 + ], + "type": "text", + "content": "[67] Yinwei Wu, Xianpan Zhou, Bing Ma, Xuefeng Su, Kai Ma, and Xinchao Wang. Ifadapter: Instance feature control for grounded text-to-image generation. arXiv:2409.08240, 2024.3" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 56, + 624, + 295, + 667 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 624, + 295, + 667 + ], + "spans": [ + { + "bbox": [ + 56, + 624, + 295, + 667 + ], + "type": "text", + "content": "[68] Jiayu Xiao, Henglei Lv, Liang Li, Shuhui Wang, and Qingming Huang. R&b: Region and boundary aware zero-shot grounded text-to-image generation. In ICLR, 2023. 2, 3, 6, 7, 5" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 56, + 670, + 295, + 712 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 670, + 295, + 712 + ], + "spans": [ + { + "bbox": [ + 56, + 670, + 295, + 712 + ], + "type": "text", + "content": "[69] Shitao Xiao, Yueze Wang, Junjie Zhou, Huaying Yuan, Xingrun Xing, Ruiran Yan, Shuting Wang, Tiejun Huang, and Zheng Liu. Omnigen: Unified image generation. arXiv:2409.11340, 2024. 3" + } + ] + } + ], + "index": 13 + } + ], + "sub_type": "ref_text" + }, + { + "bbox": [ + 316, + 73, + 554, + 677 + ], + "type": "list", + "angle": 0, + "index": 29, + "blocks": [ + { + "bbox": [ + 316, + 73, + 554, + 117 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 73, + 554, + 117 + ], + "spans": [ + { + "bbox": [ + 316, + 73, + 554, + 117 + ], + "type": "text", + "content": "[70] Jinheng Xie, Yuexiang Li, Yawen Huang, Haozhe Liu, Wentian Zhang, Yefeng Zheng, and Mike Zheng Shou. Boxdiff: Text-to-image synthesis with training-free box-constrained diffusion. In ICCV, pages 7452-7461, 2023. 2, 3, 6, 7, 5" + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 316, + 118, + 554, + 161 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 118, + 554, + 161 + ], + "spans": [ + { + "bbox": [ + 316, + 118, + 554, + 161 + ], + "type": "text", + "content": "[71] Xingqian Xu, Jiayi Guo, Zhangyang Wang, Gao Huang, Irfan Essa, and Humphrey Shi. Prompt-free diffusion: Taking\" text\" out of text-to-image diffusion models. In CVPR, pages 8682-8692, 2024. 3" + } + ] + } + ], + "index": 16 + }, + { + "bbox": [ + 316, + 163, + 554, + 206 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 163, + 554, + 206 + ], + "spans": [ + { + "bbox": [ + 316, + 163, + 554, + 206 + ], + "type": "text", + "content": "[72] Binxin Yang, Shuyang Gu, Bo Zhang, Ting Zhang, Xuejin Chen, Xiaoyan Sun, Dong Chen, and Fang Wen. Paint by example: Exemplar-based image editing with diffusion models. In CVPR, pages 18381-18391, 2023. 3, 1, 6" + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 316, + 208, + 554, + 251 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 208, + 554, + 251 + ], + "spans": [ + { + "bbox": [ + 316, + 208, + 554, + 251 + ], + "type": "text", + "content": "[73] Ling Yang, Zhaochen Yu, Chenlin Meng, Minkai Xu, Stefano Ermon, and CUI Bin. Mastering text-to-image diffusion: Recaptioning, planning, and generating with multimodal llms. In ICML, 2024. 3, 6, 7, 5" + } + ] + } + ], + "index": 18 + }, + { + "bbox": [ + 316, + 252, + 554, + 296 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 252, + 554, + 296 + ], + "spans": [ + { + "bbox": [ + 316, + 252, + 554, + 296 + ], + "type": "text", + "content": "[74] Ling Yang, Zhilong Zhang, Zhaochen Yu, Jingwei Liu, Minkai Xu, Stefano Ermon, and Bin Cui. Cross-modal contextualized diffusion models for text-guided visual generation and editing. arXiv:2402.16627, 2024. 3" + } + ] + } + ], + "index": 19 + }, + { + "bbox": [ + 316, + 297, + 554, + 351 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 297, + 554, + 351 + ], + "spans": [ + { + "bbox": [ + 316, + 297, + 554, + 351 + ], + "type": "text", + "content": "[75] Yang Yang, Wen Wang, Liang Peng, Chaotian Song, Yao Chen, Hengjia Li, Xiaolong Yang, Qinglin Lu, Deng Cai, Boxi Wu, et al. Lora-composer: Leveraging low-rank adaptation for multi-concept customization in training-free diffusion models. arXiv:2403.11627, 2024. 1" + } + ] + } + ], + "index": 20 + }, + { + "bbox": [ + 316, + 353, + 554, + 397 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 353, + 554, + 397 + ], + "spans": [ + { + "bbox": [ + 316, + 353, + 554, + 397 + ], + "type": "text", + "content": "[76] Zhengyuan Yang, Jianfeng Wang, Zhe Gan, Linjie Li, Kevin Lin, Chenfei Wu, Nan Duan, Zicheng Liu, Ce Liu, Michael Zeng, et al. Reco: Region-controlled text-to-image generation. In CVPR, pages 14246-14255, 2023. 3" + } + ] + } + ], + "index": 21 + }, + { + "bbox": [ + 316, + 399, + 554, + 430 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 399, + 554, + 430 + ], + "spans": [ + { + "bbox": [ + 316, + 399, + 554, + 430 + ], + "type": "text", + "content": "[77] Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv:2308.06721, 2023. 1" + } + ] + } + ], + "index": 22 + }, + { + "bbox": [ + 316, + 432, + 554, + 475 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 432, + 554, + 475 + ], + "spans": [ + { + "bbox": [ + 316, + 432, + 554, + 475 + ], + "type": "text", + "content": "[78] Ziyang Yuan, Mingdeng Cao, Xintao Wang, Zhonggang Qi, Chun Yuan, and Ying Shan. Customnet: Zero-shot object customization with variable-viewpoints in text-to-image diffusion models. arXiv:2310.19784, 2023. 1" + } + ] + } + ], + "index": 23 + }, + { + "bbox": [ + 316, + 476, + 554, + 510 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 476, + 554, + 510 + ], + "spans": [ + { + "bbox": [ + 316, + 476, + 554, + 510 + ], + "type": "text", + "content": "[79] Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In ICCV, pages 3836-3847, 2023. 3, 6, 7, 8, 5, 9" + } + ] + } + ], + "index": 24 + }, + { + "bbox": [ + 316, + 511, + 554, + 553 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 511, + 554, + 553 + ], + "spans": [ + { + "bbox": [ + 316, + 511, + 554, + 553 + ], + "type": "text", + "content": "[80] Lirui Zhao, Tianshuo Yang, Wenqi Shao, Yuxin Zhang, Yu Qiao, Ping Luo, Kaipeng Zhang, and Rongrong Ji. Diffree: Text-guided shape free object inpainting with diffusion model. arXiv:2407.16982, 2024. 3, 1" + } + ] + } + ], + "index": 25 + }, + { + "bbox": [ + 316, + 555, + 554, + 598 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 555, + 554, + 598 + ], + "spans": [ + { + "bbox": [ + 316, + 555, + 554, + 598 + ], + "type": "text", + "content": "[81] Shihao Zhao, Dongdong Chen, Yen-Chun Chen, Jianmin Bao, Shaozhe Hao, Lu Yuan, and Kwan-Yee K Wong. Uni-controlnet: All-in-one control to text-to-image diffusion models. In NIPS, 2023. 3" + } + ] + } + ], + "index": 26 + }, + { + "bbox": [ + 316, + 601, + 554, + 633 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 601, + 554, + 633 + ], + "spans": [ + { + "bbox": [ + 316, + 601, + 554, + 633 + ], + "type": "text", + "content": "[82] Dewei Zhou, You Li, Fan Ma, Zongxin Yang, and Yi Yang. Mige " + }, + { + "bbox": [ + 316, + 601, + 554, + 633 + ], + "type": "inline_equation", + "content": "^{+ + }" + }, + { + "bbox": [ + 316, + 601, + 554, + 633 + ], + "type": "text", + "content": " : Advanced multi-instance generation controller for image synthesis. arXiv:2407.02329, 2024. 3" + } + ] + } + ], + "index": 27 + }, + { + "bbox": [ + 316, + 635, + 554, + 677 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 635, + 554, + 677 + ], + "spans": [ + { + "bbox": [ + 316, + 635, + 554, + 677 + ], + "type": "text", + "content": "[83] Dewei Zhou, You Li, Fan Ma, Xiaoting Zhang, and Yi Yang. Mige: Multi-instance generation controller for text-to-image synthesis. In CVPR, pages 6818-6828, 2024. 1, 2, 3, 6, 7, 8, 5, 9" + } + ] + } + ], + "index": 28 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 294, + 748, + 318, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 294, + 748, + 318, + 757 + ], + "spans": [ + { + "bbox": [ + 294, + 748, + 318, + 757 + ], + "type": "text", + "content": "18389" + } + ] + } + ], + "index": 30 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 10 + } + ], + "_backend": "vlm", + "_version_name": "2.6.4" +} \ No newline at end of file diff --git a/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/1135c1a4-3af0-45ac-a497-2c9557104d8c_content_list.json b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/1135c1a4-3af0-45ac-a497-2c9557104d8c_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..97671ad471e4205d449ee02607718b57d5809da5 --- /dev/null +++ b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/1135c1a4-3af0-45ac-a497-2c9557104d8c_content_list.json @@ -0,0 +1,1666 @@ +[ + { + "type": "text", + "text": "VSNet: Focusing on the Linguistic Characteristics of Sign Language", + "text_level": 1, + "bbox": [ + 153, + 130, + 843, + 152 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Yuhao Li $^{1}$ , Xinyue Chen $^{1}$ , Hongkai Li $^{1}$ , Xiaorong Pu $^{1,3}$ , Peng Jin $^{2}$ , Yazhou Ren $^{1,3,*}$ $^{1}$ School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China; $^{2}$ Sichuan Provincial Key Lab of Philosophy and Social Science for Language Intelligence in Special Education, Leshan Normal University, Leshan, China; $^{3}$ Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China, Shenzhen, China", + "bbox": [ + 156, + 179, + 864, + 271 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract", + "text_level": 1, + "bbox": [ + 246, + 304, + 326, + 319 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Sign language is a visual language expressed through complex movements of the upper body. The human skeleton plays a critical role in sign language recognition due to its good separation from the video background. However, mainstream skeleton-based sign language recognition models often overly focus on the natural connections between joints, treating sign language as ordinary human movements, which neglects its linguistic characteristics. We believe that just as letters form words, each sign language gloss can also be decomposed into smaller visual symbols. To fully harness the potential of skeleton data, this paper proposes a novel joint fusion strategy and a visual symbol attention model. Specifically, we first input the complete set of skeletal joints, and after dynamically exchanging joint information, we discard the parts with the weakest connections to other joints, resulting in a fused, simplified skeleton. Then, we group the joints most likely to express the same visual symbol and discuss the joint movements within each group separately. To validate the superiority of our method, we conduct extensive experiments on multiple public benchmark datasets. The results show that, without complex pretraining, we still achieve new state-of-the-art performance. The code is available at https://github.com/atinyboy/VSNet.", + "bbox": [ + 89, + 335, + 483, + 684 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1. Introduction", + "text_level": 1, + "bbox": [ + 91, + 729, + 220, + 744 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Sign language is a visual language expressed through body movements and serves as the primary communication tool for the deaf community [13]. Although sign language and spoken language have distinctly different grammatical structures [1, 7], they share common linguistic features [39, 44, 59]. Accurate sign language translation can facilitate communication between deaf and hearing people, fostering a more inclusive and accessible society. Among", + "bbox": [ + 89, + 753, + 482, + 876 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/0ef7164ed9cc0e33de67ef10e8993b7a4a21ae4ff844e5a166b44fb3d111af05.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 522, + 303, + 614, + 375 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/44fe15157fcbe074d6c1d7cd80d06ce1f7f18531872187c473e9164ff0c00f43.jpg", + "image_caption": [ + "gloss: 'accent'" + ], + "image_footnote": [], + "bbox": [ + 617, + 303, + 709, + 375 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/f9bf732e363bf30348c2d295f318f5d05c852e71aa7b797239453d99c7a09b96.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 710, + 303, + 803, + 375 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/ca5a47bc0d09f7265a256a59d8607956570ca3711d2ec5e38228d7f2796734ad.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 805, + 303, + 897, + 375 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/fdeac63056da7ab56e05b49bb7c6ccfa708b60992d811714809658431cc4ca36.jpg", + "image_caption": [ + "Figure 1. In the sign for \"accent\", the index and middle fingers (or only index finger) are extended toward the neck while other fingers are curled, forming the \"pointing\" VS. In the sign for \"algebra\", movements can be decomposed into multiple VS such as \"crossed arms,\" \"fist,\" and \"thumb-up.\"" + ], + "image_footnote": [], + "bbox": [ + 522, + 396, + 614, + 467 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/18fa9e3dbae1a5067929bcacce7062c5be4971a7cddd1ad711e479f11e05de4b.jpg", + "image_caption": [ + "gloss:'algebra'" + ], + "image_footnote": [], + "bbox": [ + 617, + 396, + 707, + 467 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/ad3e18e00f9ec6d67c41252c0742477cc98782276705a52e360e368fa82d5f19.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 710, + 396, + 803, + 467 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/10785b8de7ae97caba6d78a7d9b114606070865a8f7c70ff8eea76801732b85a.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 805, + 396, + 895, + 467 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "the key subtasks in sign language translation is isolated sign language recognition (ISLR), which focuses on translating individual glosses. ISLR has garnered significant attention as it aids people in accurately understanding numerous gestures and quickly learning sign language.", + "bbox": [ + 511, + 594, + 906, + 671 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "In recent years, human skeletons have been increasingly applied in the ISLR field due to their strong operability and clear separation from video backgrounds. Although most skeleton-based recognition methods have achieved promising results, they typically treat sign language as typical human motion [12, 21, 23, 30, 31, 41, 58], overlooking the intrinsic relationships among joints and the unique linguistic characteristics [25, 52] of sign language itself. We believe that: (1) The importance of joints in sign language is not equivalent; joints with stronger connections to others are more significant than isolated joints, and connections among joints with similar movement tendencies hold more meaning than those among dispersed joints. (2) Similar to how letters form words, each sign language gloss can be decomposed into smaller visual symbols (VS). While VS", + "bbox": [ + 511, + 674, + 908, + 902 + ], + "page_idx": 0 + }, + { + "type": "header", + "text": "CVF", + "bbox": [ + 106, + 2, + 181, + 42 + ], + "page_idx": 0 + }, + { + "type": "header", + "text": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore.", + "bbox": [ + 236, + 0, + 810, + 46 + ], + "page_idx": 0 + }, + { + "type": "page_footnote", + "text": "*Corresponding author (yazhou.ren@uestc.edu.cn).", + "bbox": [ + 107, + 886, + 382, + 900 + ], + "page_idx": 0 + }, + { + "type": "page_number", + "text": "24320", + "bbox": [ + 478, + 944, + 519, + 957 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "lack explicit meaning on their own, from a skeletal perspective, they capture distinct movement patterns of joints, effectively decomposing sign language actions (see Fig. 1). Therefore, our work focuses on these two core aspects:", + "bbox": [ + 89, + 90, + 482, + 151 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Core 1: Focus on identifying essential joints. Including too many joints can impede training, while too few can lead to information loss. On the one hand, many prior models have used manually trimmed, simplified skeletons, but manually selecting the optimal joints is not trivial. On the other hand, enabling a model to automatically choose key joints can be computationally intensive. Fortunately, the dynamically learned skeletons from GCNs approximate real structures more closely, enabling us to reduce the dependency on an initial skeleton. By computing the weight of each joint, we can identify and retain the most important joints while discarding less critical local joints. Specifically, for the complete hand skeleton, we discard a small number of joints based on computed weights after each GCN layer, relearning new skeleton connections in the subsequent GCN layer. We control the number of joints discarded within each region to ensure that the final GCN output retains a recognizable structure. Notably, even joints with the lowest weights undergo at least one convolution, so their information is not entirely discarded.", + "bbox": [ + 91, + 154, + 482, + 455 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Core 2: Focus on emphasizing visual symbols (VS) in sign language. If each joint is considered the smallest element of a sign language action, then the combination of joints across all frames becomes too vast to compute self-attention. In practice, sign language is not random motion; it is designed for human recognition. When a signer interprets sign language, they do not scrutinize the position of every joint but instead form a general impression of each hand. Thus, in this work, we concentrate on recognizing VS. Unfortunately, isolating each VS from movements is as challenging as the entire SLR task itself. Therefore, we adopt a simplified approach, segmenting the skeleton into a generalized VS in a fixed manner, as shown in Fig. 2. We then design two self-attention modules to capture spatial and temporal relationships between these VS. Empirically, we find that this approach significantly enhances the model's discriminative power.", + "bbox": [ + 91, + 458, + 482, + 715 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Our main contributions are summarized as follows:", + "bbox": [ + 109, + 717, + 449, + 731 + ], + "page_idx": 1 + }, + { + "type": "list", + "sub_type": "text", + "list_items": [ + "- We propose a novel module with an automated skeleton simplification function for sign language recognition. This module is designed to dynamically learn global skeleton connections while progressively discarding isolated local joints.", + "- Considering the linguistic characteristics of sign language, we introduce a new self-attention model called VSformer, which captures temporal and spatial patterns in sign language motions and extracts knowledge from the human skeleton based on the concept of VS.", + "- We conduct extensive experiments on ISLR to validate" + ], + "bbox": [ + 89, + 734, + 482, + 900 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/643f5b9a8f8824d4be18635ddea1b3f7e73bb10c6cc91a7c3c93649d5a9787aa.jpg", + "image_caption": [ + "Figure 2. Decomposing generalized visual symbols from videos" + ], + "image_footnote": [], + "bbox": [ + 522, + 90, + 893, + 256 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "the effectiveness of our proposed approach. Compared to previous methods, our approach achieves SoTA performance on four public benchmarks, namely WLASL [32], MSASL [26], NMFs-CSR [16], and Slovo [27].", + "bbox": [ + 524, + 310, + 903, + 371 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2. Related Works", + "text_level": 1, + "bbox": [ + 513, + 383, + 663, + 398 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2.1. Sign Language Recognition", + "text_level": 1, + "bbox": [ + 513, + 409, + 759, + 425 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Sign language is the primary means of communication within the deaf community. With significant advances in machine learning and computer vision over the past decade, exploring sign language recognition (SLR) has become increasingly important [28, 43, 45], as it can automatically interpret sign language and help deaf and mute individuals communicate more easily in daily life. Unlike directly translating sign language sentences, isolated sign language recognition (ISLR) is a subtask in the field of sign language translation [45], aimed at translating each individual gloss [40, 53]. ISLR is not only useful for guiding beginners in learning sign language and assisting in the creation of new sign language gestures, but it also lays a solid foundation for SLR [1, 4, 53]. Research in this area can be divided into two categories based on the input modality: RGB-based methods and skeleton-based methods.", + "bbox": [ + 511, + 431, + 906, + 672 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "RGB-based Recognition. RGB-based methods, similar to other video recognition tasks, typically use CNNs to extract visual features from RGB videos [5, 15, 19, 20, 29, 55-57], and then aggregate temporal and spatial information to predict sign language. However, RGB-based models face two significant challenges: first, they are heavily affected by changes in the video background, and second, video data occupies a large amount of memory resources.", + "bbox": [ + 511, + 674, + 906, + 794 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Skeleton-based Recognition. Compared to RGB images, human skeletons [8, 47, 50, 54] describe the motion information conveyed in sign language videos using simplified points and edges, reducing interference from complex backgrounds and the appearance of the signer. Moreover, operations on three-dimensional spatial points with actual physical meanings offer stronger interpretability and", + "bbox": [ + 511, + 795, + 906, + 901 + ], + "page_idx": 1 + }, + { + "type": "page_number", + "text": "24321", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/a0fe65ed0df5a97890f903093b93c51daf44c0b86f7de44230e2f011d4ddd927.jpg", + "image_caption": [ + "Figure 3. Overview of our VSNet. Here, Down sample consists of a convolution layer with a kernel size of 7 and a stride of 2 along the temporal dimension, followed by a batch norm layer, Average Pooling represent transforming the data into a single-dimensional feature by taking the mean across both temporal and spatial dimensions. For details on the Skeleton- $(a,b,c)$ , please refer to Fig. 4." + ], + "image_footnote": [], + "bbox": [ + 99, + 88, + 890, + 345 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "lower memory consumption. In earlier works, identifying the correct human keypoints in a video was a major challenge. However, with the development of deep learning, human pose estimation models have become increasingly reliable [10, 24, 51].", + "bbox": [ + 89, + 422, + 483, + 500 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.2. Human Skeleton", + "text_level": 1, + "bbox": [ + 89, + 511, + 256, + 526 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Due to the natural physical connections between skeleton joints, Graph Convolutional Network (GCN) has been widely applied to model spatio-temporal relationships [12, 23, 31, 34, 41, 46]. GCN require predefined connectivity of the key points to construct the graph network. However, the physically natural skeleton connections may not always be the most suitable edges for recognition. Relying solely on the skeleton itself to construct the graph can limit the information flow between nodes. To address this issue, SL-GCN [21] employs a dynamic graph with decoupling between different frames, while TD-GCN [37] adds supernodes to control the connections between nodes. A common drawback of these approaches is that they require complex networks to model the key points. Furthermore, while dynamic graphs alleviate the limitations of the original skeleton, blindly allowing all key points to connect introduces redundant information, leading to suboptimal models.", + "bbox": [ + 88, + 535, + 482, + 792 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Therefore, we argue that relying solely on GCN does not fully exploit the true potential of skeleton data.", + "bbox": [ + 89, + 792, + 483, + 824 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In recent years, attention-based methods [3, 6, 9, 38, 42, 48] have also made progress in human skeleton recognition. Unlike GCN, attention mechanisms do not require the researcher to provide information on joint connections, instead equally attending to all nodes. However, captur", + "bbox": [ + 89, + 825, + 483, + 901 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "ing the correlations between all joints across all frames requires substantial memory resources. To address this, SLGTformer [49] decomposes the spatio-temporal pose sequence into spatial graphs and temporal windows, while SkateFormer [11] partitions joints and frames and performs self-attention on each partition. Based on these theories, we design a model where GCNs and self-attention networks complement each other. GCNs are responsible for exploring the spatial correlations between human joints, while self-attention networks focus on capturing the motion of the skeleton across multiple time sequences and spatial combinations.", + "bbox": [ + 511, + 422, + 906, + 603 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3. Methods", + "text_level": 1, + "bbox": [ + 513, + 622, + 612, + 636 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.1. Data Pre-Processing", + "text_level": 1, + "bbox": [ + 511, + 647, + 705, + 665 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Key Point Extraction. Sign language is a visual language primarily conveyed through hand and body movements, with facial expressions also playing a crucial role. However, the human skeleton mainly describes bones and joints, excluding muscles, making it unable to accurately capture facial expressions in sign language videos. Our investigation revealed that a significant number of signers, especially beginners, do not exhibit extensive facial expressions during sign language communication. Therefore, our main experiments exclude facial expressions, while the appendix provides additional details on how the concept of VS can be extended to the face.", + "bbox": [ + 511, + 672, + 906, + 852 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In previous GCN models, finger joints were often simplified to avoid making the graph too complex, leading to a loss of information. In this study, we selected a complete", + "bbox": [ + 511, + 854, + 908, + 902 + ], + "page_idx": 2 + }, + { + "type": "page_number", + "text": "24322", + "bbox": [ + 478, + 944, + 519, + 957 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/3b302e9367d27a69b446d6fd883fe21af9ef8d9b70365c49823d0a11074b393e.jpg", + "image_caption": [ + "Figure 4. Illustration of the skeleton and grouping type-1. $V_{0}$ represents a localization node, the dashed box indicates nodes that are grouped together and hollow dots indicate dropped nodes. Please note that (a) is the input real skeleton diagram, while (b) and (c) are simplified representations drawn after dropping for ease of analysis." + ], + "image_footnote": [], + "bbox": [ + 94, + 90, + 584, + 215 + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/b715aea289b6efb25c9f8e0f180402f2a4f181c1c5aefe14eb1470b89cdc6b34.jpg", + "image_caption": [ + "Figure 5. Skeleton groupings type-2 and type-3. The other hand is grouped in a mirrored fashion." + ], + "image_footnote": [], + "bbox": [ + 620, + 90, + 898, + 234 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "set of 44 key points from both hands out of the 133 human body joints extracted using HRNet [24, 51], as shown in Fig. 4(a).", + "bbox": [ + 88, + 311, + 480, + 357 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Data Enhancing. Given a video $X \\in \\mathbb{R}^{T \\times V \\times C}$ , where T, V, and C represent the number of frames, joints, and coordinates respectively. We replace abnormal points—such as those out of range (less than 0, more than 512) or clearly distorted—caused by model extraction errors, with the average value from adjacent frames. Additionally, to standardize body positions and movement amplitudes across different signers in the sign language videos, we first use the hollow node $V_{0}$ in Fig. 4 as the reference node for the entire video. Then, all key points in each frame are shifted based on this reference coordinate, as follows:", + "bbox": [ + 88, + 357, + 482, + 523 + ], + "page_idx": 3 + }, + { + "type": "equation", + "text": "\n$$\nX (t, v, c) = X (t, v, c) - X (t, v _ {0}, c) + [ x, y, z ] \\tag {1}\n$$\n", + "text_format": "latex", + "bbox": [ + 130, + 539, + 482, + 556 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $[x,y,z]$ is the average value of all the coordinates of $V_{0}$ in the dataset.", + "bbox": [ + 89, + 569, + 483, + 599 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In the time dimension, we randomly select 64 frames to standardize the video length and perform random mirroring and shifting of the video to generate augmented data.", + "bbox": [ + 89, + 599, + 483, + 646 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.2. Self-Pacing Dropping Block (SPD)", + "text_level": 1, + "bbox": [ + 89, + 657, + 390, + 675 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "As shown in Fig. 3, our SPD consists of two components: \"Keypoint Activation\" and \"Weak Joints Dropping\". The input is a complete hand skeleton where the node information is independent, and the output is a high-dimensional feature skeleton after activation. The entire block is stacked three times, where the third block only outputs features and does not build a new skeleton.", + "bbox": [ + 89, + 683, + 482, + 787 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Keypoint Activation Layer. In this part, we define the interaction of each joint with others as an activation. We employ a decoupled graph convolution [21] to perform this activation, represented as:", + "bbox": [ + 89, + 789, + 483, + 849 + ], + "page_idx": 3 + }, + { + "type": "equation", + "text": "\n$$\nF _ {o u t} ^ {t} = \\frac {1}{p} \\sum_ {k = 0} ^ {p} D ^ {- \\frac {1}{2}} A _ {k} D ^ {- \\frac {1}{2}} \\phi \\left(F _ {i n} ^ {t}\\right) \\tag {2}\n$$\n", + "text_format": "latex", + "bbox": [ + 166, + 864, + 482, + 904 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $A$ is the learnable adjacency matrix initialized as $I + In + Out$ , $D$ presents the diagonal degree of $A$ . Here, $I$ denotes self-connections, while $In$ and $Out$ represent inward and outward connections, respectively. The symbol $\\phi$ represents a linear layer, and $p$ represents the decoupling quantity, with a different $A$ applied each time, followed by averaging the results. Following the graph convolution, we apply the simplest Temporal Convolutional Network (TCN) to capture features along the temporal dimension.", + "bbox": [ + 511, + 311, + 906, + 449 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Weak Joints Dropping Layer. Compared to the simplified skeletons used in [21, 49], our approach utilizes a complete set of skeleton keypoints that contains richer information. However, this comes at the cost of significantly increasing the model's burden and raising the risk of introducing outlier points. To leverage these additional points with minimal overhead, we have designed an adaptive skeleton simplification strategy.", + "bbox": [ + 511, + 452, + 908, + 575 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Before formally introducing this layer, it is necessary to present our approach to skeleton decomposition. Based on the definition of VS described in Introduction, the decomposition concept involves grouping joints that exhibit strong overall correlation or frequently form VS together in sign language. These groupings can be viewed as a generalized or macro-level VS. A simple approach is grouping each finger and arm separately, but this creates too many groups, potentially affecting later interactions. Based on this, we conducted various experiments on VS, including treating individual fingers and distant fingers as VS. However, these experiments did not yield optimal results (the average performance in WLASL dataset decreased by $1.20\\%$ and $1.64\\%$ , respectively). We believe this is because the information from a single finger is insufficient to form a meaningful VS, while the information disparity between distant fingers is even greater. This also serves as the skeleton grouping basis for Figs. 4 and 5.", + "bbox": [ + 511, + 578, + 908, + 851 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "After graph convolution in Eq. 2, $A(i,j)$ has learned the weighted edge between node $i$ and node $j$ . These edges reflect the connection strength between $i$ and $j$ . Since these", + "bbox": [ + 511, + 854, + 908, + 902 + ], + "page_idx": 3 + }, + { + "type": "page_number", + "text": "24323", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "connections are not normalized via softmax, we can use $A$ as a guide to calculate the total correlation of all keypoints, Cor, as follows:", + "bbox": [ + 89, + 90, + 483, + 135 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nC o r _ {i} = \\sum_ {j} \\max \\left(A _ {0} (i, j), A _ {1}, \\dots , A _ {p}\\right) \\tag {3}\n$$\n", + "text_format": "latex", + "bbox": [ + 153, + 148, + 482, + 180 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "A lower $Cor$ to other keypoints indicates that the point is more independent. We refer to the two most independent points in each group as weak joints, which are dropped from the skeleton. To ensure that the number of keypoints in each group remains consistent, the groups involving both arms will have two fewer joints dropped.", + "bbox": [ + 89, + 184, + 483, + 275 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.3.VSformer", + "text_level": 1, + "bbox": [ + 89, + 282, + 205, + 297 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "After the joint fusion block, we obtained six different joint groupings, as shown in Fig. 4(c). Based on the concept of skeletal disassembly, we treat each grouping as a generalized VS. We now need to focus on two main questions: Q1, what are the relationships between different VS? And Q2, do these VS remain consistent throughout the entire video, or do they change after a brief period?", + "bbox": [ + 89, + 306, + 483, + 411 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Inspired by [11], which proposes an efficient model for partitioning human limbs in action recognition, we employ two types of self-attention to constrain the model's focus on the aforementioned questions. Specifically, we calculate spatial pairing attention among different combinations within a frame to address Q1, and we compute temporal window attention for both global and local motion of the combinations across different frames to address Q2.", + "bbox": [ + 89, + 412, + 483, + 531 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Spatial Pairing Attention (SPA). Given a frame $F_{t} \\in \\mathbb{R}^{G \\times C}$ , where $G = 6$ represents VS groupings, the ideal approach would involve calculating attention among all VS pairwise. However, this is inefficient and often purposeless, as VS that are far apart in the skeletal structure tend to have weak correlations. For instance, in the sign language gesture for \"accent\", it is unnecessary to compare the fingers of the left hand with those of the unused right hand. Based on this observation, we ultimately partition $F_{t}$ into pairs as follows:", + "bbox": [ + 89, + 532, + 483, + 681 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nF _ {t} = \\left[ P _ {\\text {a r m}}, P _ {\\text {l e f t}}, P _ {\\text {r i g h t}} \\right] \\tag {4}\n$$\n", + "text_format": "latex", + "bbox": [ + 209, + 683, + 480, + 700 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $P_{\\mathrm{arm}} \\in \\mathbb{R}^{2 \\times C}$ represents the grouping of both arms, $P_{\\mathrm{left}}$ denotes the four fingers of the left hand, and $P_{\\mathrm{right}}$ represents the remaining four fingers of the right hand. Represented as:", + "bbox": [ + 89, + 704, + 483, + 763 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nS P A _ {\\text {o u t}} = \\operatorname {c o n c a t} \\left(\\operatorname {s o f t m a x} \\left(\\frac {Q _ {i} K _ {i} ^ {T}}{\\sqrt {h}}\\right) V _ {i}\\right) \\tag {5}\n$$\n", + "text_format": "latex", + "bbox": [ + 140, + 768, + 482, + 804 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $Q_{i}, K_{i}, V_{i}$ are obtained from $P_{i}$ after mapping through linear layers. The attention is actually calculated among the 8 joints within each pairing and is ultimately concatenated to form the output.", + "bbox": [ + 89, + 809, + 482, + 869 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Temporal Window Attention (TWA). Depending on different sign language videos, the VS can be classified into", + "bbox": [ + 89, + 869, + 483, + 900 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/621111fb7906cca097896dddac1e940b4f52be4a0b9723b1c4add3959b904881.jpg", + "image_caption": [ + "Figure 6. TWA windows. The small ticks represent frames, and the large ticks represent windows." + ], + "image_footnote": [], + "bbox": [ + 524, + 95, + 903, + 135 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "two types: short duration and long duration, representing their respective duration periods. Accordingly, our TWA is also divided into two types: TWA-s and TWA-l. Given a video $X \\in \\mathbb{R}^{T \\times G \\times C}$ , where $T$ represents the total number of frames, the timeline is divided into $n$ windows, each containing $k$ frames, such that $T = n \\times k$ , as shown in Fig. 6. For TWA-s, attention is computed within each window. For TWA-l, frames with the same index are selected sequentially from each window. Finally, the results from both types of attention are concatenated to produce the output, represented as:", + "bbox": [ + 511, + 208, + 906, + 375 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nT W A _ {o u t} = \\operatorname {s o f t m a x} \\left(\\phi_ {1} (W) \\phi_ {2} ^ {T} (W) + b\\right) \\phi_ {3} (W) \\tag {6}\n$$\n", + "text_format": "latex", + "bbox": [ + 535, + 385, + 906, + 404 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $W$ represents the selected windows, $\\phi_{n}$ denotes dividing the $C$ dimension into three equal parts and selecting the $n$ -th part, and $b$ represents the added bias.", + "bbox": [ + 511, + 414, + 905, + 458 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "It is worth noting that the discussions on Q1 and Q2 are not fully addressed by these two attention layers. The SPA only addresses the relationships between paired VS, while the windows defined in TWA do not entirely correspond to the duration of the VS. Therefore, we incorporate a simple GCN and TCN to enhance the model's ability to address the remaining scenarios. The connection method of SPA and TWA is shown in the right side of Fig. 3. After passing the input data $X \\in \\mathbb{R}^{T \\times G \\times C}$ through a linear layer, we divide its channel dimension $C$ into four equal parts, sending each part into one of the four modules and later concatenated. Additionally, unlike the Decoupled GCN in Sect. 3.2, the GCN here utilizes an adjacency matrix shared across certain frames, which significantly reduces the computational load.", + "bbox": [ + 511, + 460, + 906, + 672 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.4. Multi-Grouping Ensemble", + "text_level": 1, + "bbox": [ + 511, + 680, + 754, + 696 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In the previous sections, we introduced the concept of skeletal disassembly based on VS. However, since we cannot precisely determine the true VS composition for each sign language, using a fixed disassembly method appears to be a compromise, potentially weakening interpretability. Therefore, we adopt an ensemble strategy to reduce such discrepancies. Specifically, we designed two additional skeleton grouping schemes, as shown in the Fig. 5, and trained the model separately to obtain the output $q$ from the final fully connected layer for each scheme. Following [21], we aggregate these results using a weighted sum, represented as: $q_{\\mathrm{total}} = \\alpha_1q_{\\mathrm{type - 1}} + \\alpha_2q_{\\mathrm{type - 2}} + \\alpha_3q_{\\mathrm{type - 3}} + \\alpha_4q_{\\mathrm{bone}}$ , where $\\alpha$ represents the weights. For a detailed discussion on $\\alpha$ ,", + "bbox": [ + 511, + 704, + 906, + 900 + ], + "page_idx": 4 + }, + { + "type": "page_number", + "text": "24324", + "bbox": [ + 478, + 944, + 519, + 955 + ], + "page_idx": 4 + }, + { + "type": "table", + "img_path": "images/70acf9951ff5aaf6d8c7cd56e115a61e787330cd2da58e5baa16e5c647a0f337.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
MethodsMSASL1000MSASL200MSASL100
P-IP-CP-IP-CP-IP-C
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
RGB-based
HMA [15]49.1669.7546.2768.6066.3084.0367.4784.0373.4589.7074.5989.70
I3D [5]*--57.6981.05--81.9793.79--81.7695.16
BSL [2]64.7185.5961.5584.43--------
TCK [33]†----80.3191.8281.1492.2483.0493.4683.9193.52
skeleton-based
ST-GCN [58]36.0359.9232.3257.1552.9176.6754.2077.6259.8482.0360.7982.96
BEST [60]58.8281.1854.8780.0576.6091.5476.7591.9580.9895.1181.2495.44
SignBERT [14]†59.8081.8657.0680.9477.3491.1078.0291.4881.3793.6682.3193.76
SignBERT+ [17]†62.4283.4960.1582.4478.51-79.35-84.94-85.23-
MASA [61]†63.4783.8960.7983.2979.2592.8679.7093.3383.2295.2483.1995.46
SSRL [62]†65.2285.0962.6884.3882.9394.7883.4395.0386.2696.9686.6396.79
Ours70.7186.3668.3885.7184.6294.0485.1994.0886.2695.2486.6895.39
Ours(4-crops)73.1888.6470.8887.5986.1795.5186.6495.6188.1197.0988.3297.18
", + "bbox": [ + 91, + 88, + 906, + 342 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/948ee5102452f7798c718f2b06fb8bcf7dfdc78e35f33a87fe17a339fff5a94e.jpg", + "table_caption": [ + "Table 1. Comparison with previous work on MSASL. “†” indicates that the model was pre-trained, and “*” indicates methods that applied result fusion with other modalities in post-processing. Bold indicates the best performance, while underlined denotes the second-best." + ], + "table_footnote": [], + "table_body": "
MethodsWLASL2000WLASL300WLASL100
P-IP-CP-IP-CP-IP-C
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
RGB-based
I3D [5]*32.4857.31--56.1479.9456.2478.3865.8984.1167.0184.58
HMA [15]37.9171.2635.9070.00--------
BSL [2]46.8279.3644.7278.47--------
TCK [33]†----68.5689.5268.7589.4177.5291.0877.5591.42
NLA-SLR V64 [63]*51.1583.4348.1482.20--------
skeleton-based
P3D [30]44.4779.6942.1878.5267.1889.0167.6289.2476.7191.9778.2792.97
BEST [60]46.2579.3343.5277.6567.6689.2268.3189.5777.9191.4777.8392.50
SignBERT [14]†47.4683.3245.1782.3270.3688.9271.1789.3679.0793.8080.0594.17
MASA [61]†49.0682.9046.9181.8073.6591.7774.3392.1383.7293.8084.4794.30
NLA-SLR K64 [63]*49.1082.0046.1880.71--------
SL-GCN [22]*51.5084.9448.8784.02--------
DSTA [18]53.68-51.17-79.97-80.56-82.38-83.09-
Ours53.5483.1851.1882.0379.0494.0179.7494.2584.5094.5785.2594.33
Ours(4-crops)55.9887.0753.5486.0480.0994.7680.8594.9285.6694.9686.2595.08
", + "bbox": [ + 91, + 393, + 906, + 670 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Table 2. Comparison with previous work on WLASL. “†” indicates that the model was pre-trained, and “*” indicates methods that applied result fusion with other modalities in post-processing. Bold indicates the best performance, while underlined denotes the second-best.", + "bbox": [ + 89, + 680, + 906, + 709 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "please refer to Sect. 4.4. The term bone refers to the joint data of type-1 represented in vector form, generated by directing from the source joint to its target joint according to the natural connections of the human body.", + "bbox": [ + 89, + 734, + 482, + 796 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4. Experiments", + "text_level": 1, + "bbox": [ + 89, + 806, + 223, + 824 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.1. Parameter Settings", + "text_level": 1, + "bbox": [ + 89, + 832, + 274, + 848 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We employ the AdamW optimizer with a plateau learning rate schedule. The initial learning rate is set to $1 \\times 10^{-3}$ , with a minimum learning rate of $1 \\times 10^{-5}$ and a decay rate", + "bbox": [ + 89, + 854, + 483, + 902 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "of 0.5. The batch size is 32 and the loss function used is cross-entropy with label smoothing. Training spans a maximum of 200 epochs, with the first 15 epochs serving as a warm-up period starting from a learning rate of $1 \\times 10^{-7}$ . A random sample of 64 frames is selected as input, and with a $50\\%$ probability, image mirroring is applied. A random jitter in the range $(-10, 10)$ is added to each joint point. In the last two layers of SPD, joints are randomly masked with a probability of $(1 - p_{kr}) / 100 \\times \\text{epoch}$ , where $p_{kr}$ is set to 0.9. All experiments were performed with NVIDIA A100.", + "bbox": [ + 511, + 734, + 906, + 887 + ], + "page_idx": 5 + }, + { + "type": "page_number", + "text": "24325", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/4b0b34d9395710f0472923dc42a5987e8754994ed5e4212f2acabb0bcd27597a.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
MethodsNMFs-CSR
Top-1Top-5
RGB-based
I3D [5]*64.488.0
HMA [15]64.791.0
GLE-Net [36]69.088.1
skeleton-based
ST-GCN [58]59.986.8
BEST [60]68.594.4
MASA [61]†71.797.0
SignBERT [14]†74.993.2
Ours75.395.4
Ours(4-crops)76.797.2
", + "bbox": [ + 161, + 88, + 415, + 289 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/6e0863849cf0b9d8091a93d7d3cd76aaf00e4eafdbdb07ae6c335cc5f9120a23.jpg", + "table_caption": [ + "Table 3. Comparison on NMFs-CSR. Bold indicates the best performance, while underlined denotes the second-best." + ], + "table_footnote": [], + "table_body": "
MethodsSlovo
Top-1Top-5
Swin-large-48 [38]55.66-
MViTv2-small-48 [35]62.18-
MViTv2-small-32 [35]64.09-
Ours76.5393.08
Ours(4-crops)77.8395.17
", + "bbox": [ + 142, + 342, + 434, + 454 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Table 4. Comparison with the baselines on Slovo. The final number indicates the number of frames input to the model.", + "bbox": [ + 89, + 462, + 482, + 493 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Metrics. Following [22, 61, 63], we report top-1 and top-5 classification accuracy, as well as per-instance (P-I) and per-class (P-C) precision metrics. Each table includes results for single-crop and 4-crop inference, with all ablation studies conducted in the single-crop setting. For 4-crops inference, we sum the predictions from the three types of skeleton groupings along with the bone data to obtain the final prediction.", + "bbox": [ + 89, + 518, + 483, + 640 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.2. Dataset", + "text_level": 1, + "bbox": [ + 89, + 650, + 187, + 664 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We conduct experiments on four publicly available benchmarks: WLASL, MSASL, NMFs-CSL, and Slovo. All training occurs on the training set, with hyperparameter tuning performed on the validation set.", + "bbox": [ + 89, + 672, + 482, + 733 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "MSASL [26] is a widely used dataset for American Sign Language (ASL). It contains a total of 16,054, 5,287, and 4,172 samples in the training, validation, and test sets, respectively, with a vocabulary of 1,000 signs. Two subsets, consisting of the top 100 and 300 words, are defined as MSASL100 and MSASL300, respectively.", + "bbox": [ + 89, + 734, + 482, + 824 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "WLASL [32] has a larger vocabulary but fewer total samples than MSASL. It contains 2,000 signs performed by over 100 signers, with 14,289, 3,916, and 2,878 samples in the training, validation, and test sets. Similarly, two subsets, WLASL100 and WLASL300, are created. Both", + "bbox": [ + 89, + 824, + 483, + 900 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "ASL datasets are collected from real-world scenarios, offering greater authenticity.", + "bbox": [ + 511, + 90, + 903, + 121 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "NMFs-CSR [16] is a challenging Chinese Sign Language (CSL) dataset with a vocabulary of 1,067 and consists of 25,608 and 6,402 samples in the training and test sets, respectively.", + "bbox": [ + 511, + 122, + 903, + 181 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Slovo [27] is the latest dataset for Russian Sign Language (RSL). It contains 20,000 videos across 1,000 isolated RSL gestures from 194 signers. Notably, the authors provide 42 hand landmarks based on MediaPipe, so we do not re-extract joint points.", + "bbox": [ + 511, + 183, + 906, + 258 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.3. Comparison with State-of-the-art Methods", + "text_level": 1, + "bbox": [ + 511, + 270, + 877, + 286 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In this section, we compare our model with several state-of-the-art approaches. For single-modality methods, we categorize them as either RGB-based or skeleton-based. For multi-modality methods, we report the performance of their single-modality models before ensemble.", + "bbox": [ + 511, + 294, + 903, + 369 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "MSASL [26]. As shown in the Tab. 1, RGB-based methods have developed slowly in recent years, and our approach shows a significant improvement when compared to them. Among skeleton-based methods, SignBERT [14] and BEST [60] designed self-supervised learning strategies to pre-train the model in order to alleviate issues related to limited interpretability and overfitting due to the small size of the SLR datasets. NASA [61] and SSRL [62] address the problem of neglecting explicit motion information and lexical signs during pre-training. However, none of these methods take into account the linguistic characteristics of sign language itself. Compared to them, our method achieves a substantial top-1 improvement of $7.96\\%$ in P-I and $8.2\\%$ in P-C on the full dataset. However, SSRL performs slightly better than our single-crop method on the 100-class subset, which we speculate is due to SPD requiring a certain amount of data to effectively learn the droppable nodes.", + "bbox": [ + 511, + 369, + 906, + 627 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "WLASL [32]. SAM-SLR [22] and NLA-SLR [63] are currently the two best-performing heavy multi-modality ensemble models. Among them, SL-GCN is the skeleton model used in SAM-SLR for later ensemble, employing a multi-stream approach to further improve performance. K64 and V64 are skeleton and RGB models in NLASLR that use 64-frame input, where the skeleton is represented as a heatmap sequence, allowing their keypoint encoder to share the same architecture with the video encoder. DSTA [18] was the previously most accurate model. Compared to their results, our top-1 accuracy achieved a $2.30\\%$ improvement on WLASL2000.", + "bbox": [ + 511, + 628, + 905, + 808 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "NMFs-CSL [16]. As shown in Tab. 3, NMFs-CSL is a balanced dataset in terms of class distribution, so we only report the P-I accuracy. The pretraining strategy of SignBERT [14] is highly effective on this dataset; however, its performance on top-5 is suboptimal. Our model does not use any additional training data, yet it shows a significant", + "bbox": [ + 511, + 809, + 905, + 900 + ], + "page_idx": 6 + }, + { + "type": "page_number", + "text": "24326", + "bbox": [ + 478, + 944, + 519, + 957 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/b906ec28865617bbba0152fb78ccecece5a76f63ed3253cd0e0fe93ef268ec63.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
SkeletonMSASL1000MSASL200MSASL100WLASL2000WLASL300WLASL100NMFs-CLR
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
type-170.7186.3684.6294.0486.2695.2453.5483.1879.0494.0184.5094.9675.2695.47
bone68.4185.8882.4193.3884.6894.0650.7682.4574.7091.1784.1193.8073.6394.31
type-270.3586.1085.2894.1187.4595.6453.0283.3278.4493.4183.7293.8074.7094.86
type-370.3586.1084.9294.1185.6095.1152.2682.4978.1493.4184.8893.4174.8795.55
", + "bbox": [ + 102, + 88, + 893, + 172 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/31e604748199a8431fe8b374c23dd2cd44acb63279e6389e154223e2aed8487b.jpg", + "table_caption": [ + "Table 5. Performance of multi-grouping VSformer on three datasets." + ], + "table_footnote": [], + "table_body": "
weightingWLASL2000MSASL1000NMFs-CSL
Top-1Top-5Top-1Top-5Top-1Top-5
sum55.9887.0773.1888.6476.6597.17
acc56.0187.1873.1388.6476.6597.17
", + "bbox": [ + 94, + 220, + 477, + 277 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/a7028c8dd42754b2060fb0519da73a576752b18f71243ebf83ae6ff3357910fb.jpg", + "table_caption": [ + "Table 6. Performance of the two ensemble methods" + ], + "table_footnote": [], + "table_body": "
MethodsWLASL300WLASL2000
Top-1Top-5Top-1Top-5
Linear74.7092.8146.3277.24
GCN76.3591.7751.1880.82
SPD(ours)79.0494.0153.5483.18
", + "bbox": [ + 124, + 316, + 449, + 396 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/a41391df11b191da55a27851e8377608ec051734a9a098a54cd740bb7ff4d18e.jpg", + "table_caption": [ + "Table 7. Comparing our SPD with GCN and linear layers." + ], + "table_footnote": [], + "table_body": "
SPATMAWLASL300WLASL2000
Top-1Top-5Top-1Top-5
✓77.6993.4152.4382.24
✓77.9993.2652.3681.83
✓✓79.0494.0153.5483.18
", + "bbox": [ + 117, + 435, + 457, + 515 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Table 8. Ablation study on the impact of our two types of attention mechanisms.", + "bbox": [ + 89, + 525, + 482, + 554 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "improvement over SignBERT in overall performance, with a $1.8\\%$ improvement in top-1 accuracy and a $4\\%$ improvement in top-5 accuracy.", + "bbox": [ + 89, + 580, + 482, + 625 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Slovo [27]. Slovo is a recently released RSL dataset, with only limited experimental data available. We provide an up-to-date benchmark for better evaluating this dataset. As shown in Tab. 4, the results for Swin [38] and MViTv2 [35] were provided as baselines by the dataset authors. Notably, the skeleton landmarks provided by the authors includes only 42 hand joints, without arm data, which partially limits our model's performance. Despite this, we still achieve results significantly above the baselines.", + "bbox": [ + 89, + 626, + 482, + 762 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.4. Ablation Study", + "text_level": 1, + "bbox": [ + 89, + 771, + 243, + 787 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In this section, we conduct several key ablation experiments on the WLASL dataset [32] and report P-I accuracy to validate the effectiveness of our approach.", + "bbox": [ + 89, + 794, + 482, + 840 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "SPD. We first replace SPD with two alternative methods. As shown in Tab. 7, \"Linear\" denotes a three-layer Conv-2D module with the same output dimension as SPD, while \"GCN\" refers to three decoupled graph convolution", + "bbox": [ + 89, + 839, + 483, + 901 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "layer without the weak joints dropping strategy. To ensure consistent output data shape, these two methods directly input the simplified 24 joint points. The complete SFB implementation achieves the highest accuracy, indicating that our dropping strategy better facilitates the extraction of sign language action features.", + "bbox": [ + 511, + 224, + 903, + 314 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Multi-Grouping. As shown in Tab. 5, we present the final performance of four types of crops. The overall performance of each grouping type varies only slightly, mainly depending on the sample distribution of the dataset. However, integrating all four crops significantly enhances action classification performance. The performance of the bone-based grouping is slightly lower than that of joint-based results, primarily because our model is designed for points and is less sensitive to vector-based data.", + "bbox": [ + 511, + 316, + 905, + 452 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "SAM-SLR [21] ultimately using manually set weights to control the ensemble. However, we argue that such weights lack interpretability. As shown in Tab. 6, we employed two methods to integrate the results: \"acc\" refers to using the individual top-1 accuracy of each crop as the weight, while \"sum\" represents directly summing the outputs of the final layer of the model. Both methods resulted in an about $2\\%$ improvement in top-1 accuracy, demonstrating that our ensemble approach is both general and robust.", + "bbox": [ + 511, + 455, + 905, + 592 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "VSformer. For our two types of self-attention, we replace one with a linear layer to test the performance of the other individually. As shown in Tab. 8, the full VSformer achieves significantly better performance, indicating that features in both temporal and spatial dimensions are crucial for distinguishing sign language actions.", + "bbox": [ + 511, + 594, + 905, + 686 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5. Conclusion", + "text_level": 1, + "bbox": [ + 511, + 705, + 633, + 722 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In this work, we design a skeleton recognition model tailored specifically for ISLR, termed VSNet. The core idea of our approach is to focus on the truly essential elements of sign language. To this end, we first introduce a weak joints dropping strategy to prioritize more critical points in the skeleton. We then propose the VSformer to capture the linguistic characteristics of sign language, where Visual Symbols (VS) represent smaller components of sign language actions. We conduct extensive experiments to validate the effectiveness of the proposed VSNet, achieving state-of-the-art performance across four benchmarks.", + "bbox": [ + 511, + 734, + 905, + 900 + ], + "page_idx": 7 + }, + { + "type": "page_number", + "text": "24327", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Acknowledgements", + "text_level": 1, + "bbox": [ + 91, + 90, + 258, + 107 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "This work was supported in part by National Natural Science Foundation of China (No. 62476052), Sichuan Science and Technology Program (No. 2024NSFSC1473), the open project of Sichuan Provincial Key Laboratory of Philosophy and Social Science for Language Intelligence in Special Education (No. YYZN-2023-3), the Humanities and Social Sciences Project of the Ministry of Education (No. 23YJA740013), and Central Guidance for Local Science and Technology Development Fund Projects (No. 2024ZYD0268).", + "bbox": [ + 89, + 116, + 485, + 267 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "References", + "text_level": 1, + "bbox": [ + 91, + 282, + 187, + 299 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[1] Nikolas Adaloglou, Theocharis Chatzis, Ilias Papastratis, Andreas Stergioulas, Georgios Th Papadopoulos, Vassia Zacharopoulou, George J Xydopoulos, Klimnis Atzakas, Dimitris Papazachariou, and Petros Daras. A comprehensive study on deep learning-based methods for sign language recognition. IEEE TMM, 24:1750-1762, 2021. 1, 2", + "[2] Samuel Albanie, Gül Varol, Liliane Momeni, Triantafyllos Afouras, Joon Son Chung, Neil Fox, and Andrew Zisserman. Bsl-1k: Scaling up co-articulated sign language recognition using mouthing cues. In ECCV. Springer, 2020. 6", + "[3] Gedas Bertasius, Heng Wang, and Lorenzo Torresani. Is space-time attention all you need for video understanding? In ICML, page 4, 2021. 3", + "[4] Amanda Cardoso Duarte, Samuel Albanie, Xavier Giro Nieto, and Gül Varol. Sign language video retrieval with freeform textual queries. In CVPR, pages 14074-14084, 2022. 2", + "[5] Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. In CVPR, pages 6299-6308, 2017. 2, 6, 7", + "[6] Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. In CVPR, pages 6299-6308, 2017. 3", + "[7] Yutong Chen, Fangyun Wei, Xiao Sun, Zhirong Wu, and Stephen Lin. A simple multi-modality transfer learning baseline for sign language translation. In CVPR, pages 5120-5130, 2022. 1", + "[8] Ke Cheng, Yifan Zhang, Congqi Cao, Lei Shi, Jian Cheng, and Hanqing Lu. Decoupling gcn with dropgraph module for skeleton-based action recognition. In ECCV, pages 536-553. Springer, 2020. 2", + "[9] Xiangxiang Chu, Zhi Tian, Yuqing Wang, Bo Zhang, Haibing Ren, Xiaolin Wei, Huaxia Xia, and Chunhua Shen. Twins: Revisiting the design of spatial attention in vision transformers. NeurIPS, 34:9355-9366, 2021. 3", + "[10] MMPose Contributors. Openmmlab pose estimation toolbox and benchmark. https://github.com/open-mmlab/mmpose, 2020.3", + "[11] Jeonghyeok Do and Munchurl Kim. Skateformer: Skeletal-temporal transformer for human action recognition. arXiv preprint arXiv:2403.09508, 2024. 3, 5" + ], + "bbox": [ + 93, + 309, + 480, + 901 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[12] Yong Du, Wei Wang, and Liang Wang. Hierarchical recurrent neural network for skeleton based action recognition. In CVPR, pages 1110-1118, 2015. 1, 3", + "[13] Karen Emmorey. Language, cognition, and the brain: Insights from sign language research. Psychology Press, 2001. 1", + "[14] Hezhen Hu, Weichao Zhao, Wengang Zhou, Yuechen Wang, and Houqiang Li. Signbert: Pre-training of hand-model-aware representation for sign language recognition. In ICCV, pages 11087-11096, 2021. 6, 7", + "[15] Hezhen Hu, Wengang Zhou, and Houqiang Li. Hand-model-aware sign language recognition. In AAAI, pages 1558-1566, 2021. 2, 6, 7", + "[16] Hezhen Hu, Wengang Zhou, Junfu Pu, and Houqiang Li. Global-local enhancement network for nmf-aware sign language recognition. ACM transactions on multimedia computing, communications, and applications (TOMM), 17(3): 1-19, 2021. 2, 7", + "[17] Hezhen Hu, Weichao Zhao, Wengang Zhou, and Houqiang Li. Signbert+: Hand-model-aware self-supervised pretraining for sign language understanding. IEEE TPAMI, 45 (9):11221-11239, 2023. 6", + "[18] Lianyu Hu, Liqing Gao, Zekang Liu, and Wei Feng. Dynamic spatial-temporal aggregation for skeleton-aware sign language recognition. In LREC/COLING, 2024. 6, 7", + "[19] Jie Huang, Wengang Zhou, Houqiang Li, and Weiping Li. Attention-based 3d-cnns for large-vocabulary sign language recognition. IEEE TCSVT, 29(9):2822-2832, 2018. 2", + "[20] Jie Huang, Wengang Zhou, Qilin Zhang, Houqiang Li, and Weiping Li. Video-based sign language recognition without temporal segmentation. In AAAI, 2018. 2", + "[21] Songyao Jiang, Bin Sun, Lichen Wang, Yue Bai, Kunpeng Li, and Yun Fu. Skeleton aware multi-modal sign language recognition. In CVPR, pages 3413-3423, 2021. 1, 3, 4, 5, 8", + "[22] Songyao Jiang, Bin Sun, Lichen Wang, Yue Bai, Kunpeng Li, and Yun Fu. Sign language recognition via skeleton-aware multi-model ensemble. arXiv preprint arXiv:2110.06161, 2021. 6, 7", + "[23] Tao Jiang, Necati Cihan Camgoz, and Richard Bowden. Skeleton: Skeletal transformers for robust body-posed estimation. In CVPR, pages 3394–3402, 2021. 1, 3", + "[24] Sheng Jin, Lumin Xu, Jin Xu, Can Wang, Wentao Liu, Chen Qian, Wanli Ouyang, and Ping Luo. Whole-body human pose estimation in the wild. In ECCV, pages 196-214. Springer, 2020. 3, 4", + "[25] Trevor Johnston and Adam Schembri. Australian Sign Language (Auslan): An introduction to sign language linguistics. Cambridge University Press, 2007. 1", + "[26] Hamid Reza Vaezi Joze and Oscar Koller. Ms-asl: A large-scale data set and benchmark for understanding american sign language. arXiv preprint arXiv:1812.01053, 2018. 2, 7", + "[27] Alexander Kapitanov, Kvanchiani Karina, Alexander Nagaev, and Petrova Elizaveta. Slovo: Russian sign language dataset. In International Conference on Computer Vision Systems, pages 63-73. Springer, 2023. 2, 7, 8" + ], + "bbox": [ + 516, + 92, + 903, + 901 + ], + "page_idx": 8 + }, + { + "type": "page_number", + "text": "24328", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[28] Oscar Koller. Quantitative survey of the state of the art in sign language recognition. arXiv preprint arXiv:2008.09918, 2020. 2", + "[29] Oscar Koller, Sepehr Zargaran, Hermann Ney, and Richard Bowden. Deep sign: Enabling robust statistical continuous sign language recognition via hybrid cnn-hmms. *IJCV*, 126: 1311–1325, 2018. 2", + "[30] Taeryung Lee, Yeonguk Oh, and Kyoung Mu Lee. Human part-wise 3d motion context learning for sign language recognition. In ICCV, pages 20740-20750, 2023. 1, 6", + "[31] Chao Li, Qiaoyong Zhong, Di Xie, and Shiliang Pu. Cooccurrence feature learning from skeleton data for action recognition and detection with hierarchical aggregation. arXiv preprint arXiv:1804.06055, 2018. 1, 3", + "[32] Dongxu Li, Cristian Rodriguez, Xin Yu, and Hongdong Li. Word-level deep sign language recognition from video: A new large-scale dataset and methods comparison. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 1459-1469, 2020. 2, 7, 8", + "[33] Dongxu Li, Xin Yu, Chenchen Xu, Lars Petersson, and Hongdong Li. Transferring cross-domain knowledge for video sign language recognition. In CVPR, pages 6205-6214, 2020. 6", + "[34] Maosen Li, Siheng Chen, Xu Chen, Ya Zhang, Yanfeng Wang, and Qi Tian. Actional-structural graph convolutional networks for skeleton-based action recognition. In CVPR, pages 3595–3603, 2019. 3", + "[35] Yanghao Li, Chao-Yuan Wu, Haoqi Fan, Karttikeya Mangalam, Bo Xiong, Jitendra Malik, and Christoph Feichtenhofer. Mviv2: Improved multiscale vision transformers for classification and detection. In CVPR, pages 4804-4814, 2022. 7, 8", + "[36] Jiajia Liao, Yujun Liu, Yingchao Piao, Jinhe Su, Guorong Cai, and Yundong Wu. Gle-net: A global and local ensemble network for aerial object detection. International Journal of Computational Intelligence Systems, 15(1):2, 2022. 7", + "[37] Jinfu Liu, Xinshun Wang, Can Wang, Yuan Gao, and Mengyuan Liu. Temporal decoupling graph convolutional network for skeleton-based gesture recognition. IEEE TMM, 26:811-823, 2023. 3", + "[38] Ze Liu, Jia Ning, Yue Cao, Yixuan Wei, Zheng Zhang, Stephen Lin, and Han Hu. Video swim transformer. In CVPR, pages 3202-3211, 2022. 3, 7, 8", + "[39] Anna Mindess. Reading between the signs: Intercultural communication for sign language interpreters. Nicholas Brealey, 2014. 1", + "[40] Liliane Momeni, Hannah Bull, KR Prajwal, Samuel Albanie, Gül Varol, and Andrew Zisserman. Automatic dense annotation of large-vocabulary sign language videos. In ECCV, pages 671-690. Springer, 2022. 2", + "[41] Evonne Ng, Shiry Ginosar, Trevor Darrell, and Hanbyul Joo. Body2hands: Learning to infer 3d hands from conversational gesture body dynamics. In CVPR, pages 11865-11874, 2021. 1, 3", + "[42] Ladislav Rampášek, Michael Galkin, Vijay Prakash Dwivedi, Anh Tuan Luu, Guy Wolf, and Dominique Beaini. Recipe for a general, powerful, scalable graph transformer. NeurIPS, 35:14501-14515, 2022. 3" + ], + "bbox": [ + 91, + 90, + 480, + 898 + ], + "page_idx": 9 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[43] Razieh Rastgoo, Kourosh Kiani, and Sergio Escalera. Sign language recognition: A deep survey. Expert Systems with Applications, 164:113794, 2021. 2", + "[44] Wendy Sandler. Sign language and linguistic universals. Cambridge University, 2006. 1", + "[45] Noha Sarhan and Simone Frintrop. Unraveling a decade: a comprehensive survey on isolated sign language recognition. In ICCV, pages 3210-3219, 2023. 2", + "[46] Lei Shi, Yifan Zhang, Jian Cheng, and Hanqing Lu. Two-stream adaptive graph convolutional networks for skeleton-based action recognition. In CVPR, pages 12026-12035, 2019. 3", + "[47] Lei Shi, Yifan Zhang, Jian Cheng, and Hanqing Lu. Skeleton-based action recognition with multi-stream adaptive graph convolutional networks. IEEE TIP, 29:9532-9545, 2020. 2", + "[48] Chenyang Si, Wentao Chen, Wei Wang, Liang Wang, and Tieniu Tan. An attention enhanced graph convolutional LSTM network for skeleton-based action recognition. In CVPR, pages 1227-1236, 2019. 3", + "[49] Neil Song and Yu Xiang. Slgtformer: An attention-based approach to sign language recognition. arXiv preprint arXiv:2212.10746, 2022. 3, 4", + "[50] Yi-Fan Song, Zhang Zhang, Caifeng Shan, and Liang Wang. Stronger, faster and more explainable: A graph convolutional baseline for skeleton-based action recognition. In proceedings of the 28th ACM international conference on multimedia, pages 1625-1633, 2020. 2", + "[51] Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. Deep high-resolution representation learning for human pose estimation. In CVPR, pages 5693-5703, 2019. 3, 4", + "[52] Clayton Valli. Linguistics of American Sign Language: An Introduction. Gallaudet University Press, 2000. 1", + "[53] Gül Varol, Liliane Momeni, Samuel Albanie, Triantafyllos Afouras, and Andrew Zisserman. Scaling up sign spotting through sign language dictionaries. *IJCV*, 130(6):1416–1439, 2022. 2", + "[54] Hongsong Wang and Liang Wang. Modeling temporal dynamics and spatial configurations of actions using two-stream recurrent neural networks. In CVPR, pages 499-508, 2017. 2", + "[55] Hanjie Wang, Xiujuan Chai, and Xin Chen. A novel sign language recognition framework using hierarchical grassmann covariance matrix. IEEE TMM, 21(11):2806-2814, 2019. 2", + "[56] Chengcheng Wei, Jian Zhao, Wengang Zhou, and Houqiang Li. Semantic boundary detection with reinforcement learning for continuous sign language recognition. IEEE TCSVT, 31(3):1138-1149, 2020.", + "[57] Pan Xie, Mengyi Zhao, and Xiaohui Hu. Pisltrc: Position-informed sign language transformer with content-aware convolution. IEEE TMM, 24:3908-3919, 2021. 2", + "[58] Sijie Yan, Yuanjun Xiong, and Dahua Lin. Spatial temporal graph convolutional networks for skeleton-based action recognition. In AAAI, 2018. 1, 6, 7", + "[59] Aoxiong Yin, Zhou Zhao, Weike Jin, Meng Zhang, Xingshan Zeng, and Xiaofei He. Mlslt: Towards multilingual sign language translation. In CVPR, pages 5109-5119, 2022. 1" + ], + "bbox": [ + 514, + 90, + 903, + 900 + ], + "page_idx": 9 + }, + { + "type": "page_number", + "text": "24329", + "bbox": [ + 478, + 944, + 519, + 955 + ], + "page_idx": 9 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[60] Weichao Zhao, Hezhen Hu, Wengang Zhou, Jiaxin Shi, and Houqiang Li. Best: Bert pre-training for sign language recognition with coupling tokenization. In AAAI, pages 3597-3605, 2023. 6, 7", + "[61] Weichao Zhao, Hezhen Hu, Wengang Zhou, Yunyao Mao, Min Wang, and Houjiang Li. Masa: Motion-aware masked autoencoder with semantic alignment for sign language recognition. IEEE TCSVT, 2024. 6, 7", + "[62] Weichao Zhao, Wengang Zhou, Hezhen Hu, Min Wang, and Houqiang Li. Self-supervised representation learning with spatial-temporal consistency for sign language recognition. arXiv preprint arXiv:2406.10501, 2024. 6, 7", + "[63] Ronglai Zuo, Fangyun Wei, and Brian Mak. Natural language-assisted sign language recognition. In CVPR, pages 14890-14900, 2023. 6, 7" + ], + "bbox": [ + 91, + 90, + 480, + 303 + ], + "page_idx": 10 + }, + { + "type": "page_number", + "text": "24330", + "bbox": [ + 478, + 945, + 517, + 955 + ], + "page_idx": 10 + } +] \ No newline at end of file diff --git a/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/1135c1a4-3af0-45ac-a497-2c9557104d8c_model.json b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/1135c1a4-3af0-45ac-a497-2c9557104d8c_model.json new file mode 100644 index 0000000000000000000000000000000000000000..3ace0602be8a395d0af368f5416044f6d044fdd4 --- /dev/null +++ b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/1135c1a4-3af0-45ac-a497-2c9557104d8c_model.json @@ -0,0 +1,2378 @@ +[ + [ + { + "type": "header", + "bbox": [ + 0.107, + 0.003, + 0.182, + 0.043 + ], + "angle": 0, + "content": "CVF" + }, + { + "type": "header", + "bbox": [ + 0.237, + 0.001, + 0.812, + 0.047 + ], + "angle": 0, + "content": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore." + }, + { + "type": "title", + "bbox": [ + 0.155, + 0.131, + 0.844, + 0.154 + ], + "angle": 0, + "content": "VSNet: Focusing on the Linguistic Characteristics of Sign Language" + }, + { + "type": "text", + "bbox": [ + 0.158, + 0.18, + 0.865, + 0.272 + ], + "angle": 0, + "content": "Yuhao Li\\(^{1}\\), Xinyue Chen\\(^{1}\\), Hongkai Li\\(^{1}\\), Xiaorong Pu\\(^{1,3}\\), Peng Jin\\(^{2}\\), Yazhou Ren\\(^{1,3,*}\\) \n\\(^{1}\\)School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China; \\(^{2}\\)Sichuan Provincial Key Lab of Philosophy and Social Science for Language Intelligence in Special Education, Leshan Normal University, Leshan, China; \\(^{3}\\)Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China, Shenzhen, China" + }, + { + "type": "title", + "bbox": [ + 0.248, + 0.305, + 0.327, + 0.32 + ], + "angle": 0, + "content": "Abstract" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.337, + 0.485, + 0.685 + ], + "angle": 0, + "content": "Sign language is a visual language expressed through complex movements of the upper body. The human skeleton plays a critical role in sign language recognition due to its good separation from the video background. However, mainstream skeleton-based sign language recognition models often overly focus on the natural connections between joints, treating sign language as ordinary human movements, which neglects its linguistic characteristics. We believe that just as letters form words, each sign language gloss can also be decomposed into smaller visual symbols. To fully harness the potential of skeleton data, this paper proposes a novel joint fusion strategy and a visual symbol attention model. Specifically, we first input the complete set of skeletal joints, and after dynamically exchanging joint information, we discard the parts with the weakest connections to other joints, resulting in a fused, simplified skeleton. Then, we group the joints most likely to express the same visual symbol and discuss the joint movements within each group separately. To validate the superiority of our method, we conduct extensive experiments on multiple public benchmark datasets. The results show that, without complex pretraining, we still achieve new state-of-the-art performance. The code is available at https://github.com/atinyboy/VSNet." + }, + { + "type": "title", + "bbox": [ + 0.092, + 0.73, + 0.222, + 0.745 + ], + "angle": 0, + "content": "1. Introduction" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.755, + 0.483, + 0.877 + ], + "angle": 0, + "content": "Sign language is a visual language expressed through body movements and serves as the primary communication tool for the deaf community [13]. Although sign language and spoken language have distinctly different grammatical structures [1, 7], they share common linguistic features [39, 44, 59]. Accurate sign language translation can facilitate communication between deaf and hearing people, fostering a more inclusive and accessible society. Among" + }, + { + "type": "image", + "bbox": [ + 0.524, + 0.304, + 0.616, + 0.375 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.618, + 0.304, + 0.71, + 0.375 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.712, + 0.304, + 0.804, + 0.375 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.806, + 0.304, + 0.898, + 0.375 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.681, + 0.378, + 0.746, + 0.388 + ], + "angle": 0, + "content": "gloss: 'accent'" + }, + { + "type": "image", + "bbox": [ + 0.524, + 0.397, + 0.616, + 0.468 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.618, + 0.397, + 0.709, + 0.468 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.712, + 0.397, + 0.804, + 0.468 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.807, + 0.397, + 0.897, + 0.468 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.68, + 0.47, + 0.745, + 0.48 + ], + "angle": 0, + "content": "gloss:'algebra'" + }, + { + "type": "image_caption", + "bbox": [ + 0.512, + 0.491, + 0.908, + 0.561 + ], + "angle": 0, + "content": "Figure 1. In the sign for \"accent\", the index and middle fingers (or only index finger) are extended toward the neck while other fingers are curled, forming the \"pointing\" VS. In the sign for \"algebra\", movements can be decomposed into multiple VS such as \"crossed arms,\" \"fist,\" and \"thumb-up.\"" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.595, + 0.907, + 0.672 + ], + "angle": 0, + "content": "the key subtasks in sign language translation is isolated sign language recognition (ISLR), which focuses on translating individual glosses. ISLR has garnered significant attention as it aids people in accurately understanding numerous gestures and quickly learning sign language." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.675, + 0.909, + 0.903 + ], + "angle": 0, + "content": "In recent years, human skeletons have been increasingly applied in the ISLR field due to their strong operability and clear separation from video backgrounds. Although most skeleton-based recognition methods have achieved promising results, they typically treat sign language as typical human motion [12, 21, 23, 30, 31, 41, 58], overlooking the intrinsic relationships among joints and the unique linguistic characteristics [25, 52] of sign language itself. We believe that: (1) The importance of joints in sign language is not equivalent; joints with stronger connections to others are more significant than isolated joints, and connections among joints with similar movement tendencies hold more meaning than those among dispersed joints. (2) Similar to how letters form words, each sign language gloss can be decomposed into smaller visual symbols (VS). While VS" + }, + { + "type": "page_footnote", + "bbox": [ + 0.109, + 0.887, + 0.383, + 0.901 + ], + "angle": 0, + "content": "*Corresponding author (yazhou.ren@uestc.edu.cn)." + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.521, + 0.958 + ], + "angle": 0, + "content": "24320" + } + ], + [ + { + "type": "text", + "bbox": [ + 0.09, + 0.092, + 0.483, + 0.152 + ], + "angle": 0, + "content": "lack explicit meaning on their own, from a skeletal perspective, they capture distinct movement patterns of joints, effectively decomposing sign language actions (see Fig. 1). Therefore, our work focuses on these two core aspects:" + }, + { + "type": "text", + "bbox": [ + 0.093, + 0.155, + 0.483, + 0.456 + ], + "angle": 0, + "content": "Core 1: Focus on identifying essential joints. Including too many joints can impede training, while too few can lead to information loss. On the one hand, many prior models have used manually trimmed, simplified skeletons, but manually selecting the optimal joints is not trivial. On the other hand, enabling a model to automatically choose key joints can be computationally intensive. Fortunately, the dynamically learned skeletons from GCNs approximate real structures more closely, enabling us to reduce the dependency on an initial skeleton. By computing the weight of each joint, we can identify and retain the most important joints while discarding less critical local joints. Specifically, for the complete hand skeleton, we discard a small number of joints based on computed weights after each GCN layer, relearning new skeleton connections in the subsequent GCN layer. We control the number of joints discarded within each region to ensure that the final GCN output retains a recognizable structure. Notably, even joints with the lowest weights undergo at least one convolution, so their information is not entirely discarded." + }, + { + "type": "text", + "bbox": [ + 0.093, + 0.459, + 0.483, + 0.716 + ], + "angle": 0, + "content": "Core 2: Focus on emphasizing visual symbols (VS) in sign language. If each joint is considered the smallest element of a sign language action, then the combination of joints across all frames becomes too vast to compute self-attention. In practice, sign language is not random motion; it is designed for human recognition. When a signer interprets sign language, they do not scrutinize the position of every joint but instead form a general impression of each hand. Thus, in this work, we concentrate on recognizing VS. Unfortunately, isolating each VS from movements is as challenging as the entire SLR task itself. Therefore, we adopt a simplified approach, segmenting the skeleton into a generalized VS in a fixed manner, as shown in Fig. 2. We then design two self-attention modules to capture spatial and temporal relationships between these VS. Empirically, we find that this approach significantly enhances the model's discriminative power." + }, + { + "type": "text", + "bbox": [ + 0.11, + 0.718, + 0.45, + 0.732 + ], + "angle": 0, + "content": "Our main contributions are summarized as follows:" + }, + { + "type": "text", + "bbox": [ + 0.091, + 0.735, + 0.483, + 0.81 + ], + "angle": 0, + "content": "- We propose a novel module with an automated skeleton simplification function for sign language recognition. This module is designed to dynamically learn global skeleton connections while progressively discarding isolated local joints." + }, + { + "type": "text", + "bbox": [ + 0.091, + 0.811, + 0.483, + 0.884 + ], + "angle": 0, + "content": "- Considering the linguistic characteristics of sign language, we introduce a new self-attention model called VSformer, which captures temporal and spatial patterns in sign language motions and extracts knowledge from the human skeleton based on the concept of VS." + }, + { + "type": "text", + "bbox": [ + 0.091, + 0.886, + 0.483, + 0.901 + ], + "angle": 0, + "content": "- We conduct extensive experiments on ISLR to validate" + }, + { + "type": "list", + "bbox": [ + 0.091, + 0.735, + 0.483, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.524, + 0.091, + 0.895, + 0.257 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.519, + 0.27, + 0.9, + 0.284 + ], + "angle": 0, + "content": "Figure 2. Decomposing generalized visual symbols from videos" + }, + { + "type": "text", + "bbox": [ + 0.526, + 0.311, + 0.905, + 0.372 + ], + "angle": 0, + "content": "the effectiveness of our proposed approach. Compared to previous methods, our approach achieves SoTA performance on four public benchmarks, namely WLASL [32], MSASL [26], NMFs-CSR [16], and Slovo [27]." + }, + { + "type": "title", + "bbox": [ + 0.514, + 0.385, + 0.664, + 0.4 + ], + "angle": 0, + "content": "2. Related Works" + }, + { + "type": "title", + "bbox": [ + 0.514, + 0.41, + 0.761, + 0.426 + ], + "angle": 0, + "content": "2.1. Sign Language Recognition" + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.433, + 0.907, + 0.674 + ], + "angle": 0, + "content": "Sign language is the primary means of communication within the deaf community. With significant advances in machine learning and computer vision over the past decade, exploring sign language recognition (SLR) has become increasingly important [28, 43, 45], as it can automatically interpret sign language and help deaf and mute individuals communicate more easily in daily life. Unlike directly translating sign language sentences, isolated sign language recognition (ISLR) is a subtask in the field of sign language translation [45], aimed at translating each individual gloss [40, 53]. ISLR is not only useful for guiding beginners in learning sign language and assisting in the creation of new sign language gestures, but it also lays a solid foundation for SLR [1, 4, 53]. Research in this area can be divided into two categories based on the input modality: RGB-based methods and skeleton-based methods." + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.675, + 0.907, + 0.795 + ], + "angle": 0, + "content": "RGB-based Recognition. RGB-based methods, similar to other video recognition tasks, typically use CNNs to extract visual features from RGB videos [5, 15, 19, 20, 29, 55-57], and then aggregate temporal and spatial information to predict sign language. However, RGB-based models face two significant challenges: first, they are heavily affected by changes in the video background, and second, video data occupies a large amount of memory resources." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.796, + 0.907, + 0.902 + ], + "angle": 0, + "content": "Skeleton-based Recognition. Compared to RGB images, human skeletons [8, 47, 50, 54] describe the motion information conveyed in sign language videos using simplified points and edges, reducing interference from complex backgrounds and the appearance of the signer. Moreover, operations on three-dimensional spatial points with actual physical meanings offer stronger interpretability and" + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.518, + 0.957 + ], + "angle": 0, + "content": "24321" + } + ], + [ + { + "type": "image", + "bbox": [ + 0.1, + 0.089, + 0.891, + 0.346 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.097, + 0.355, + 0.9, + 0.398 + ], + "angle": 0, + "content": "Figure 3. Overview of our VSNet. Here, Down sample consists of a convolution layer with a kernel size of 7 and a stride of 2 along the temporal dimension, followed by a batch norm layer, Average Pooling represent transforming the data into a single-dimensional feature by taking the mean across both temporal and spatial dimensions. For details on the Skeleton-\\((a,b,c)\\), please refer to Fig. 4." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.424, + 0.484, + 0.501 + ], + "angle": 0, + "content": "lower memory consumption. In earlier works, identifying the correct human keypoints in a video was a major challenge. However, with the development of deep learning, human pose estimation models have become increasingly reliable [10, 24, 51]." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.512, + 0.258, + 0.527 + ], + "angle": 0, + "content": "2.2. Human Skeleton" + }, + { + "type": "text", + "bbox": [ + 0.089, + 0.536, + 0.483, + 0.793 + ], + "angle": 0, + "content": "Due to the natural physical connections between skeleton joints, Graph Convolutional Network (GCN) has been widely applied to model spatio-temporal relationships [12, 23, 31, 34, 41, 46]. GCN require predefined connectivity of the key points to construct the graph network. However, the physically natural skeleton connections may not always be the most suitable edges for recognition. Relying solely on the skeleton itself to construct the graph can limit the information flow between nodes. To address this issue, SL-GCN [21] employs a dynamic graph with decoupling between different frames, while TD-GCN [37] adds supernodes to control the connections between nodes. A common drawback of these approaches is that they require complex networks to model the key points. Furthermore, while dynamic graphs alleviate the limitations of the original skeleton, blindly allowing all key points to connect introduces redundant information, leading to suboptimal models." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.794, + 0.484, + 0.825 + ], + "angle": 0, + "content": "Therefore, we argue that relying solely on GCN does not fully exploit the true potential of skeleton data." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.826, + 0.484, + 0.902 + ], + "angle": 0, + "content": "In recent years, attention-based methods [3, 6, 9, 38, 42, 48] have also made progress in human skeleton recognition. Unlike GCN, attention mechanisms do not require the researcher to provide information on joint connections, instead equally attending to all nodes. However, captur" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.424, + 0.908, + 0.604 + ], + "angle": 0, + "content": "ing the correlations between all joints across all frames requires substantial memory resources. To address this, SLGTformer [49] decomposes the spatio-temporal pose sequence into spatial graphs and temporal windows, while SkateFormer [11] partitions joints and frames and performs self-attention on each partition. Based on these theories, we design a model where GCNs and self-attention networks complement each other. GCNs are responsible for exploring the spatial correlations between human joints, while self-attention networks focus on capturing the motion of the skeleton across multiple time sequences and spatial combinations." + }, + { + "type": "title", + "bbox": [ + 0.514, + 0.623, + 0.613, + 0.637 + ], + "angle": 0, + "content": "3. Methods" + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.648, + 0.707, + 0.666 + ], + "angle": 0, + "content": "3.1. Data Pre-Processing" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.673, + 0.907, + 0.853 + ], + "angle": 0, + "content": "Key Point Extraction. Sign language is a visual language primarily conveyed through hand and body movements, with facial expressions also playing a crucial role. However, the human skeleton mainly describes bones and joints, excluding muscles, making it unable to accurately capture facial expressions in sign language videos. Our investigation revealed that a significant number of signers, especially beginners, do not exhibit extensive facial expressions during sign language communication. Therefore, our main experiments exclude facial expressions, while the appendix provides additional details on how the concept of VS can be extended to the face." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.856, + 0.909, + 0.903 + ], + "angle": 0, + "content": "In previous GCN models, finger joints were often simplified to avoid making the graph too complex, leading to a loss of information. In this study, we selected a complete" + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.521, + 0.958 + ], + "angle": 0, + "content": "24322" + } + ], + [ + { + "type": "image", + "bbox": [ + 0.096, + 0.092, + 0.585, + 0.216 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.09, + 0.23, + 0.596, + 0.288 + ], + "angle": 0, + "content": "Figure 4. Illustration of the skeleton and grouping type-1. \\( V_{0} \\) represents a localization node, the dashed box indicates nodes that are grouped together and hollow dots indicate dropped nodes. Please note that (a) is the input real skeleton diagram, while (b) and (c) are simplified representations drawn after dropping for ease of analysis." + }, + { + "type": "image", + "bbox": [ + 0.622, + 0.092, + 0.9, + 0.235 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.616, + 0.254, + 0.907, + 0.283 + ], + "angle": 0, + "content": "Figure 5. Skeleton groupings type-2 and type-3. The other hand is grouped in a mirrored fashion." + }, + { + "type": "text", + "bbox": [ + 0.089, + 0.313, + 0.482, + 0.358 + ], + "angle": 0, + "content": "set of 44 key points from both hands out of the 133 human body joints extracted using HRNet [24, 51], as shown in Fig. 4(a)." + }, + { + "type": "text", + "bbox": [ + 0.089, + 0.358, + 0.483, + 0.525 + ], + "angle": 0, + "content": "Data Enhancing. Given a video \\( X \\in \\mathbb{R}^{T \\times V \\times C} \\), where T, V, and C represent the number of frames, joints, and coordinates respectively. We replace abnormal points—such as those out of range (less than 0, more than 512) or clearly distorted—caused by model extraction errors, with the average value from adjacent frames. Additionally, to standardize body positions and movement amplitudes across different signers in the sign language videos, we first use the hollow node \\( V_{0} \\) in Fig. 4 as the reference node for the entire video. Then, all key points in each frame are shifted based on this reference coordinate, as follows:" + }, + { + "type": "equation", + "bbox": [ + 0.131, + 0.54, + 0.483, + 0.558 + ], + "angle": 0, + "content": "\\[\nX (t, v, c) = X (t, v, c) - X (t, v _ {0}, c) + [ x, y, z ] \\tag {1}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.57, + 0.484, + 0.6 + ], + "angle": 0, + "content": "where \\([x,y,z]\\) is the average value of all the coordinates of \\(V_{0}\\) in the dataset." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.601, + 0.484, + 0.647 + ], + "angle": 0, + "content": "In the time dimension, we randomly select 64 frames to standardize the video length and perform random mirroring and shifting of the video to generate augmented data." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.659, + 0.391, + 0.676 + ], + "angle": 0, + "content": "3.2. Self-Pacing Dropping Block (SPD)" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.684, + 0.483, + 0.789 + ], + "angle": 0, + "content": "As shown in Fig. 3, our SPD consists of two components: \"Keypoint Activation\" and \"Weak Joints Dropping\". The input is a complete hand skeleton where the node information is independent, and the output is a high-dimensional feature skeleton after activation. The entire block is stacked three times, where the third block only outputs features and does not build a new skeleton." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.79, + 0.484, + 0.851 + ], + "angle": 0, + "content": "Keypoint Activation Layer. In this part, we define the interaction of each joint with others as an activation. We employ a decoupled graph convolution [21] to perform this activation, represented as:" + }, + { + "type": "equation", + "bbox": [ + 0.168, + 0.865, + 0.483, + 0.905 + ], + "angle": 0, + "content": "\\[\nF _ {o u t} ^ {t} = \\frac {1}{p} \\sum_ {k = 0} ^ {p} D ^ {- \\frac {1}{2}} A _ {k} D ^ {- \\frac {1}{2}} \\phi \\left(F _ {i n} ^ {t}\\right) \\tag {2}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.313, + 0.907, + 0.45 + ], + "angle": 0, + "content": "where \\( A \\) is the learnable adjacency matrix initialized as \\( I + In + Out \\), \\( D \\) presents the diagonal degree of \\( A \\). Here, \\( I \\) denotes self-connections, while \\( In \\) and \\( Out \\) represent inward and outward connections, respectively. The symbol \\( \\phi \\) represents a linear layer, and \\( p \\) represents the decoupling quantity, with a different \\( A \\) applied each time, followed by averaging the results. Following the graph convolution, we apply the simplest Temporal Convolutional Network (TCN) to capture features along the temporal dimension." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.453, + 0.909, + 0.576 + ], + "angle": 0, + "content": "Weak Joints Dropping Layer. Compared to the simplified skeletons used in [21, 49], our approach utilizes a complete set of skeleton keypoints that contains richer information. However, this comes at the cost of significantly increasing the model's burden and raising the risk of introducing outlier points. To leverage these additional points with minimal overhead, we have designed an adaptive skeleton simplification strategy." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.579, + 0.909, + 0.852 + ], + "angle": 0, + "content": "Before formally introducing this layer, it is necessary to present our approach to skeleton decomposition. Based on the definition of VS described in Introduction, the decomposition concept involves grouping joints that exhibit strong overall correlation or frequently form VS together in sign language. These groupings can be viewed as a generalized or macro-level VS. A simple approach is grouping each finger and arm separately, but this creates too many groups, potentially affecting later interactions. Based on this, we conducted various experiments on VS, including treating individual fingers and distant fingers as VS. However, these experiments did not yield optimal results (the average performance in WLASL dataset decreased by \\(1.20\\%\\) and \\(1.64\\%\\), respectively). We believe this is because the information from a single finger is insufficient to form a meaningful VS, while the information disparity between distant fingers is even greater. This also serves as the skeleton grouping basis for Figs. 4 and 5." + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.856, + 0.909, + 0.903 + ], + "angle": 0, + "content": "After graph convolution in Eq. 2, \\( A(i,j) \\) has learned the weighted edge between node \\( i \\) and node \\( j \\). These edges reflect the connection strength between \\( i \\) and \\( j \\). Since these" + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "24323" + } + ], + [ + { + "type": "text", + "bbox": [ + 0.09, + 0.092, + 0.485, + 0.136 + ], + "angle": 0, + "content": "connections are not normalized via softmax, we can use \\(A\\) as a guide to calculate the total correlation of all keypoints, Cor, as follows:" + }, + { + "type": "equation", + "bbox": [ + 0.154, + 0.149, + 0.483, + 0.181 + ], + "angle": 0, + "content": "\\[\nC o r _ {i} = \\sum_ {j} \\max \\left(A _ {0} (i, j), A _ {1}, \\dots , A _ {p}\\right) \\tag {3}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.185, + 0.484, + 0.276 + ], + "angle": 0, + "content": "A lower \\( Cor \\) to other keypoints indicates that the point is more independent. We refer to the two most independent points in each group as weak joints, which are dropped from the skeleton. To ensure that the number of keypoints in each group remains consistent, the groups involving both arms will have two fewer joints dropped." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.284, + 0.206, + 0.299 + ], + "angle": 0, + "content": "3.3.VSformer" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.307, + 0.484, + 0.412 + ], + "angle": 0, + "content": "After the joint fusion block, we obtained six different joint groupings, as shown in Fig. 4(c). Based on the concept of skeletal disassembly, we treat each grouping as a generalized VS. We now need to focus on two main questions: Q1, what are the relationships between different VS? And Q2, do these VS remain consistent throughout the entire video, or do they change after a brief period?" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.413, + 0.484, + 0.532 + ], + "angle": 0, + "content": "Inspired by [11], which proposes an efficient model for partitioning human limbs in action recognition, we employ two types of self-attention to constrain the model's focus on the aforementioned questions. Specifically, we calculate spatial pairing attention among different combinations within a frame to address Q1, and we compute temporal window attention for both global and local motion of the combinations across different frames to address Q2." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.533, + 0.484, + 0.682 + ], + "angle": 0, + "content": "Spatial Pairing Attention (SPA). Given a frame \\(F_{t} \\in \\mathbb{R}^{G \\times C}\\), where \\(G = 6\\) represents VS groupings, the ideal approach would involve calculating attention among all VS pairwise. However, this is inefficient and often purposeless, as VS that are far apart in the skeletal structure tend to have weak correlations. For instance, in the sign language gesture for \"accent\", it is unnecessary to compare the fingers of the left hand with those of the unused right hand. Based on this observation, we ultimately partition \\(F_{t}\\) into pairs as follows:" + }, + { + "type": "equation", + "bbox": [ + 0.21, + 0.684, + 0.482, + 0.701 + ], + "angle": 0, + "content": "\\[\nF _ {t} = \\left[ P _ {\\text {a r m}}, P _ {\\text {l e f t}}, P _ {\\text {r i g h t}} \\right] \\tag {4}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.705, + 0.484, + 0.764 + ], + "angle": 0, + "content": "where \\( P_{\\mathrm{arm}} \\in \\mathbb{R}^{2 \\times C} \\) represents the grouping of both arms, \\( P_{\\mathrm{left}} \\) denotes the four fingers of the left hand, and \\( P_{\\mathrm{right}} \\) represents the remaining four fingers of the right hand. Represented as:" + }, + { + "type": "equation", + "bbox": [ + 0.141, + 0.77, + 0.483, + 0.805 + ], + "angle": 0, + "content": "\\[\nS P A _ {\\text {o u t}} = \\operatorname {c o n c a t} \\left(\\operatorname {s o f t m a x} \\left(\\frac {Q _ {i} K _ {i} ^ {T}}{\\sqrt {h}}\\right) V _ {i}\\right) \\tag {5}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.81, + 0.483, + 0.87 + ], + "angle": 0, + "content": "where \\( Q_{i}, K_{i}, V_{i} \\) are obtained from \\( P_{i} \\) after mapping through linear layers. The attention is actually calculated among the 8 joints within each pairing and is ultimately concatenated to form the output." + }, + { + "type": "text", + "bbox": [ + 0.091, + 0.871, + 0.484, + 0.901 + ], + "angle": 0, + "content": "Temporal Window Attention (TWA). Depending on different sign language videos, the VS can be classified into" + }, + { + "type": "image", + "bbox": [ + 0.525, + 0.096, + 0.905, + 0.136 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.513, + 0.153, + 0.907, + 0.181 + ], + "angle": 0, + "content": "Figure 6. TWA windows. The small ticks represent frames, and the large ticks represent windows." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.209, + 0.907, + 0.375 + ], + "angle": 0, + "content": "two types: short duration and long duration, representing their respective duration periods. Accordingly, our TWA is also divided into two types: TWA-s and TWA-l. Given a video \\( X \\in \\mathbb{R}^{T \\times G \\times C} \\), where \\( T \\) represents the total number of frames, the timeline is divided into \\( n \\) windows, each containing \\( k \\) frames, such that \\( T = n \\times k \\), as shown in Fig. 6. For TWA-s, attention is computed within each window. For TWA-l, frames with the same index are selected sequentially from each window. Finally, the results from both types of attention are concatenated to produce the output, represented as:" + }, + { + "type": "equation", + "bbox": [ + 0.536, + 0.386, + 0.907, + 0.405 + ], + "angle": 0, + "content": "\\[\nT W A _ {o u t} = \\operatorname {s o f t m a x} \\left(\\phi_ {1} (W) \\phi_ {2} ^ {T} (W) + b\\right) \\phi_ {3} (W) \\tag {6}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.415, + 0.906, + 0.459 + ], + "angle": 0, + "content": "where \\(W\\) represents the selected windows, \\(\\phi_{n}\\) denotes dividing the \\(C\\) dimension into three equal parts and selecting the \\(n\\)-th part, and \\(b\\) represents the added bias." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.461, + 0.907, + 0.673 + ], + "angle": 0, + "content": "It is worth noting that the discussions on Q1 and Q2 are not fully addressed by these two attention layers. The SPA only addresses the relationships between paired VS, while the windows defined in TWA do not entirely correspond to the duration of the VS. Therefore, we incorporate a simple GCN and TCN to enhance the model's ability to address the remaining scenarios. The connection method of SPA and TWA is shown in the right side of Fig. 3. After passing the input data \\( X \\in \\mathbb{R}^{T \\times G \\times C} \\) through a linear layer, we divide its channel dimension \\( C \\) into four equal parts, sending each part into one of the four modules and later concatenated. Additionally, unlike the Decoupled GCN in Sect. 3.2, the GCN here utilizes an adjacency matrix shared across certain frames, which significantly reduces the computational load." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.681, + 0.755, + 0.698 + ], + "angle": 0, + "content": "3.4. Multi-Grouping Ensemble" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.705, + 0.907, + 0.901 + ], + "angle": 0, + "content": "In the previous sections, we introduced the concept of skeletal disassembly based on VS. However, since we cannot precisely determine the true VS composition for each sign language, using a fixed disassembly method appears to be a compromise, potentially weakening interpretability. Therefore, we adopt an ensemble strategy to reduce such discrepancies. Specifically, we designed two additional skeleton grouping schemes, as shown in the Fig. 5, and trained the model separately to obtain the output \\( q \\) from the final fully connected layer for each scheme. Following [21], we aggregate these results using a weighted sum, represented as: \\( q_{\\mathrm{total}} = \\alpha_1q_{\\mathrm{type - 1}} + \\alpha_2q_{\\mathrm{type - 2}} + \\alpha_3q_{\\mathrm{type - 3}} + \\alpha_4q_{\\mathrm{bone}} \\), where \\( \\alpha \\) represents the weights. For a detailed discussion on \\( \\alpha \\)," + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.521, + 0.957 + ], + "angle": 0, + "content": "24324" + } + ], + [ + { + "type": "table", + "bbox": [ + 0.093, + 0.089, + 0.907, + 0.343 + ], + "angle": 0, + "content": "
MethodsMSASL1000MSASL200MSASL100
P-IP-CP-IP-CP-IP-C
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
RGB-based
HMA [15]49.1669.7546.2768.6066.3084.0367.4784.0373.4589.7074.5989.70
I3D [5]*--57.6981.05--81.9793.79--81.7695.16
BSL [2]64.7185.5961.5584.43--------
TCK [33]†----80.3191.8281.1492.2483.0493.4683.9193.52
skeleton-based
ST-GCN [58]36.0359.9232.3257.1552.9176.6754.2077.6259.8482.0360.7982.96
BEST [60]58.8281.1854.8780.0576.6091.5476.7591.9580.9895.1181.2495.44
SignBERT [14]†59.8081.8657.0680.9477.3491.1078.0291.4881.3793.6682.3193.76
SignBERT+ [17]†62.4283.4960.1582.4478.51-79.35-84.94-85.23-
MASA [61]†63.4783.8960.7983.2979.2592.8679.7093.3383.2295.2483.1995.46
SSRL [62]†65.2285.0962.6884.3882.9394.7883.4395.0386.2696.9686.6396.79
Ours70.7186.3668.3885.7184.6294.0485.1994.0886.2695.2486.6895.39
Ours(4-crops)73.1888.6470.8887.5986.1795.5186.6495.6188.1197.0988.3297.18
" + }, + { + "type": "table_caption", + "bbox": [ + 0.09, + 0.352, + 0.907, + 0.381 + ], + "angle": 0, + "content": "Table 1. Comparison with previous work on MSASL. “†” indicates that the model was pre-trained, and “*” indicates methods that applied result fusion with other modalities in post-processing. Bold indicates the best performance, while underlined denotes the second-best." + }, + { + "type": "table", + "bbox": [ + 0.093, + 0.394, + 0.907, + 0.671 + ], + "angle": 0, + "content": "
MethodsWLASL2000WLASL300WLASL100
P-IP-CP-IP-CP-IP-C
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
RGB-based
I3D [5]*32.4857.31--56.1479.9456.2478.3865.8984.1167.0184.58
HMA [15]37.9171.2635.9070.00--------
BSL [2]46.8279.3644.7278.47--------
TCK [33]†----68.5689.5268.7589.4177.5291.0877.5591.42
NLA-SLR V64 [63]*51.1583.4348.1482.20--------
skeleton-based
P3D [30]44.4779.6942.1878.5267.1889.0167.6289.2476.7191.9778.2792.97
BEST [60]46.2579.3343.5277.6567.6689.2268.3189.5777.9191.4777.8392.50
SignBERT [14]†47.4683.3245.1782.3270.3688.9271.1789.3679.0793.8080.0594.17
MASA [61]†49.0682.9046.9181.8073.6591.7774.3392.1383.7293.8084.4794.30
NLA-SLR K64 [63]*49.1082.0046.1880.71--------
SL-GCN [22]*51.5084.9448.8784.02--------
DSTA [18]53.68-51.17-79.97-80.56-82.38-83.09-
Ours53.5483.1851.1882.0379.0494.0179.7494.2584.5094.5785.2594.33
Ours(4-crops)55.9887.0753.5486.0480.0994.7680.8594.9285.6694.9686.2595.08
" + }, + { + "type": "table_caption", + "bbox": [ + 0.09, + 0.681, + 0.907, + 0.71 + ], + "angle": 0, + "content": "Table 2. Comparison with previous work on WLASL. “†” indicates that the model was pre-trained, and “*” indicates methods that applied result fusion with other modalities in post-processing. Bold indicates the best performance, while underlined denotes the second-best." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.736, + 0.483, + 0.797 + ], + "angle": 0, + "content": "please refer to Sect. 4.4. The term bone refers to the joint data of type-1 represented in vector form, generated by directing from the source joint to its target joint according to the natural connections of the human body." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.808, + 0.225, + 0.825 + ], + "angle": 0, + "content": "4. Experiments" + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.833, + 0.276, + 0.849 + ], + "angle": 0, + "content": "4.1. Parameter Settings" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.856, + 0.484, + 0.903 + ], + "angle": 0, + "content": "We employ the AdamW optimizer with a plateau learning rate schedule. The initial learning rate is set to \\(1 \\times 10^{-3}\\), with a minimum learning rate of \\(1 \\times 10^{-5}\\) and a decay rate" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.736, + 0.907, + 0.888 + ], + "angle": 0, + "content": "of 0.5. The batch size is 32 and the loss function used is cross-entropy with label smoothing. Training spans a maximum of 200 epochs, with the first 15 epochs serving as a warm-up period starting from a learning rate of \\(1 \\times 10^{-7}\\). A random sample of 64 frames is selected as input, and with a \\(50\\%\\) probability, image mirroring is applied. A random jitter in the range \\((-10, 10)\\) is added to each joint point. In the last two layers of SPD, joints are randomly masked with a probability of \\((1 - p_{kr}) / 100 \\times \\text{epoch}\\), where \\(p_{kr}\\) is set to 0.9. All experiments were performed with NVIDIA A100." + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "24325" + } + ], + [ + { + "type": "table", + "bbox": [ + 0.163, + 0.089, + 0.416, + 0.29 + ], + "angle": 0, + "content": "
MethodsNMFs-CSR
Top-1Top-5
RGB-based
I3D [5]*64.488.0
HMA [15]64.791.0
GLE-Net [36]69.088.1
skeleton-based
ST-GCN [58]59.986.8
BEST [60]68.594.4
MASA [61]†71.797.0
SignBERT [14]†74.993.2
Ours75.395.4
Ours(4-crops)76.797.2
" + }, + { + "type": "table_caption", + "bbox": [ + 0.09, + 0.3, + 0.484, + 0.329 + ], + "angle": 0, + "content": "Table 3. Comparison on NMFs-CSR. Bold indicates the best performance, while underlined denotes the second-best." + }, + { + "type": "table", + "bbox": [ + 0.143, + 0.343, + 0.435, + 0.455 + ], + "angle": 0, + "content": "
MethodsSlovo
Top-1Top-5
Swin-large-48 [38]55.66-
MViTv2-small-48 [35]62.18-
MViTv2-small-32 [35]64.09-
Ours76.5393.08
Ours(4-crops)77.8395.17
" + }, + { + "type": "table_caption", + "bbox": [ + 0.09, + 0.463, + 0.483, + 0.494 + ], + "angle": 0, + "content": "Table 4. Comparison with the baselines on Slovo. The final number indicates the number of frames input to the model." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.519, + 0.484, + 0.641 + ], + "angle": 0, + "content": "Metrics. Following [22, 61, 63], we report top-1 and top-5 classification accuracy, as well as per-instance (P-I) and per-class (P-C) precision metrics. Each table includes results for single-crop and 4-crop inference, with all ablation studies conducted in the single-crop setting. For 4-crops inference, we sum the predictions from the three types of skeleton groupings along with the bone data to obtain the final prediction." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.651, + 0.188, + 0.665 + ], + "angle": 0, + "content": "4.2. Dataset" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.674, + 0.483, + 0.734 + ], + "angle": 0, + "content": "We conduct experiments on four publicly available benchmarks: WLASL, MSASL, NMFs-CSL, and Slovo. All training occurs on the training set, with hyperparameter tuning performed on the validation set." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.735, + 0.483, + 0.825 + ], + "angle": 0, + "content": "MSASL [26] is a widely used dataset for American Sign Language (ASL). It contains a total of 16,054, 5,287, and 4,172 samples in the training, validation, and test sets, respectively, with a vocabulary of 1,000 signs. Two subsets, consisting of the top 100 and 300 words, are defined as MSASL100 and MSASL300, respectively." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.825, + 0.484, + 0.901 + ], + "angle": 0, + "content": "WLASL [32] has a larger vocabulary but fewer total samples than MSASL. It contains 2,000 signs performed by over 100 signers, with 14,289, 3,916, and 2,878 samples in the training, validation, and test sets. Similarly, two subsets, WLASL100 and WLASL300, are created. Both" + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.092, + 0.905, + 0.122 + ], + "angle": 0, + "content": "ASL datasets are collected from real-world scenarios, offering greater authenticity." + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.123, + 0.905, + 0.183 + ], + "angle": 0, + "content": "NMFs-CSR [16] is a challenging Chinese Sign Language (CSL) dataset with a vocabulary of 1,067 and consists of 25,608 and 6,402 samples in the training and test sets, respectively." + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.184, + 0.907, + 0.26 + ], + "angle": 0, + "content": "Slovo [27] is the latest dataset for Russian Sign Language (RSL). It contains 20,000 videos across 1,000 isolated RSL gestures from 194 signers. Notably, the authors provide 42 hand landmarks based on MediaPipe, so we do not re-extract joint points." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.271, + 0.878, + 0.287 + ], + "angle": 0, + "content": "4.3. Comparison with State-of-the-art Methods" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.295, + 0.905, + 0.37 + ], + "angle": 0, + "content": "In this section, we compare our model with several state-of-the-art approaches. For single-modality methods, we categorize them as either RGB-based or skeleton-based. For multi-modality methods, we report the performance of their single-modality models before ensemble." + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.371, + 0.907, + 0.628 + ], + "angle": 0, + "content": "MSASL [26]. As shown in the Tab. 1, RGB-based methods have developed slowly in recent years, and our approach shows a significant improvement when compared to them. Among skeleton-based methods, SignBERT [14] and BEST [60] designed self-supervised learning strategies to pre-train the model in order to alleviate issues related to limited interpretability and overfitting due to the small size of the SLR datasets. NASA [61] and SSRL [62] address the problem of neglecting explicit motion information and lexical signs during pre-training. However, none of these methods take into account the linguistic characteristics of sign language itself. Compared to them, our method achieves a substantial top-1 improvement of \\(7.96\\%\\) in P-I and \\(8.2\\%\\) in P-C on the full dataset. However, SSRL performs slightly better than our single-crop method on the 100-class subset, which we speculate is due to SPD requiring a certain amount of data to effectively learn the droppable nodes." + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.629, + 0.906, + 0.809 + ], + "angle": 0, + "content": "WLASL [32]. SAM-SLR [22] and NLA-SLR [63] are currently the two best-performing heavy multi-modality ensemble models. Among them, SL-GCN is the skeleton model used in SAM-SLR for later ensemble, employing a multi-stream approach to further improve performance. K64 and V64 are skeleton and RGB models in NLASLR that use 64-frame input, where the skeleton is represented as a heatmap sequence, allowing their keypoint encoder to share the same architecture with the video encoder. DSTA [18] was the previously most accurate model. Compared to their results, our top-1 accuracy achieved a \\(2.30\\%\\) improvement on WLASL2000." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.81, + 0.906, + 0.901 + ], + "angle": 0, + "content": "NMFs-CSL [16]. As shown in Tab. 3, NMFs-CSL is a balanced dataset in terms of class distribution, so we only report the P-I accuracy. The pretraining strategy of SignBERT [14] is highly effective on this dataset; however, its performance on top-5 is suboptimal. Our model does not use any additional training data, yet it shows a significant" + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.521, + 0.958 + ], + "angle": 0, + "content": "24326" + } + ], + [ + { + "type": "table", + "bbox": [ + 0.104, + 0.089, + 0.894, + 0.173 + ], + "angle": 0, + "content": "
SkeletonMSASL1000MSASL200MSASL100WLASL2000WLASL300WLASL100NMFs-CLR
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
type-170.7186.3684.6294.0486.2695.2453.5483.1879.0494.0184.5094.9675.2695.47
bone68.4185.8882.4193.3884.6894.0650.7682.4574.7091.1784.1193.8073.6394.31
type-270.3586.1085.2894.1187.4595.6453.0283.3278.4493.4183.7293.8074.7094.86
type-370.3586.1084.9294.1185.6095.1152.2682.4978.1493.4184.8893.4174.8795.55
" + }, + { + "type": "table_caption", + "bbox": [ + 0.295, + 0.184, + 0.7, + 0.198 + ], + "angle": 0, + "content": "Table 5. Performance of multi-grouping VSformer on three datasets." + }, + { + "type": "table", + "bbox": [ + 0.095, + 0.221, + 0.478, + 0.279 + ], + "angle": 0, + "content": "
weightingWLASL2000MSASL1000NMFs-CSL
Top-1Top-5Top-1Top-5Top-1Top-5
sum55.9887.0773.1888.6476.6597.17
acc56.0187.1873.1388.6476.6597.17
" + }, + { + "type": "table_caption", + "bbox": [ + 0.134, + 0.29, + 0.441, + 0.302 + ], + "angle": 0, + "content": "Table 6. Performance of the two ensemble methods" + }, + { + "type": "table", + "bbox": [ + 0.125, + 0.317, + 0.45, + 0.397 + ], + "angle": 0, + "content": "
MethodsWLASL300WLASL2000
Top-1Top-5Top-1Top-5
Linear74.7092.8146.3277.24
GCN76.3591.7751.1880.82
SPD(ours)79.0494.0153.5483.18
" + }, + { + "type": "table_caption", + "bbox": [ + 0.115, + 0.408, + 0.458, + 0.422 + ], + "angle": 0, + "content": "Table 7. Comparing our SPD with GCN and linear layers." + }, + { + "type": "table", + "bbox": [ + 0.118, + 0.436, + 0.458, + 0.516 + ], + "angle": 0, + "content": "
SPATMAWLASL300WLASL2000
Top-1Top-5Top-1Top-5
✓77.6993.4152.4382.24
✓77.9993.2652.3681.83
✓✓79.0494.0153.5483.18
" + }, + { + "type": "table_caption", + "bbox": [ + 0.09, + 0.526, + 0.483, + 0.555 + ], + "angle": 0, + "content": "Table 8. Ablation study on the impact of our two types of attention mechanisms." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.581, + 0.483, + 0.626 + ], + "angle": 0, + "content": "improvement over SignBERT in overall performance, with a \\(1.8\\%\\) improvement in top-1 accuracy and a \\(4\\%\\) improvement in top-5 accuracy." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.627, + 0.483, + 0.763 + ], + "angle": 0, + "content": "Slovo [27]. Slovo is a recently released RSL dataset, with only limited experimental data available. We provide an up-to-date benchmark for better evaluating this dataset. As shown in Tab. 4, the results for Swin [38] and MViTv2 [35] were provided as baselines by the dataset authors. Notably, the skeleton landmarks provided by the authors includes only 42 hand joints, without arm data, which partially limits our model's performance. Despite this, we still achieve results significantly above the baselines." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.772, + 0.245, + 0.789 + ], + "angle": 0, + "content": "4.4. Ablation Study" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.795, + 0.483, + 0.841 + ], + "angle": 0, + "content": "In this section, we conduct several key ablation experiments on the WLASL dataset [32] and report P-I accuracy to validate the effectiveness of our approach." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.84, + 0.484, + 0.902 + ], + "angle": 0, + "content": "SPD. We first replace SPD with two alternative methods. As shown in Tab. 7, \"Linear\" denotes a three-layer Conv-2D module with the same output dimension as SPD, while \"GCN\" refers to three decoupled graph convolution" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.225, + 0.905, + 0.315 + ], + "angle": 0, + "content": "layer without the weak joints dropping strategy. To ensure consistent output data shape, these two methods directly input the simplified 24 joint points. The complete SFB implementation achieves the highest accuracy, indicating that our dropping strategy better facilitates the extraction of sign language action features." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.317, + 0.906, + 0.453 + ], + "angle": 0, + "content": "Multi-Grouping. As shown in Tab. 5, we present the final performance of four types of crops. The overall performance of each grouping type varies only slightly, mainly depending on the sample distribution of the dataset. However, integrating all four crops significantly enhances action classification performance. The performance of the bone-based grouping is slightly lower than that of joint-based results, primarily because our model is designed for points and is less sensitive to vector-based data." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.456, + 0.906, + 0.593 + ], + "angle": 0, + "content": "SAM-SLR [21] ultimately using manually set weights to control the ensemble. However, we argue that such weights lack interpretability. As shown in Tab. 6, we employed two methods to integrate the results: \"acc\" refers to using the individual top-1 accuracy of each crop as the weight, while \"sum\" represents directly summing the outputs of the final layer of the model. Both methods resulted in an about \\(2\\%\\) improvement in top-1 accuracy, demonstrating that our ensemble approach is both general and robust." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.595, + 0.906, + 0.687 + ], + "angle": 0, + "content": "VSformer. For our two types of self-attention, we replace one with a linear layer to test the performance of the other individually. As shown in Tab. 8, the full VSformer achieves significantly better performance, indicating that features in both temporal and spatial dimensions are crucial for distinguishing sign language actions." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.707, + 0.634, + 0.723 + ], + "angle": 0, + "content": "5. Conclusion" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.735, + 0.906, + 0.901 + ], + "angle": 0, + "content": "In this work, we design a skeleton recognition model tailored specifically for ISLR, termed VSNet. The core idea of our approach is to focus on the truly essential elements of sign language. To this end, we first introduce a weak joints dropping strategy to prioritize more critical points in the skeleton. We then propose the VSformer to capture the linguistic characteristics of sign language, where Visual Symbols (VS) represent smaller components of sign language actions. We conduct extensive experiments to validate the effectiveness of the proposed VSNet, achieving state-of-the-art performance across four benchmarks." + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "24327" + } + ], + [ + { + "type": "title", + "bbox": [ + 0.092, + 0.091, + 0.259, + 0.108 + ], + "angle": 0, + "content": "Acknowledgements" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.117, + 0.486, + 0.268 + ], + "angle": 0, + "content": "This work was supported in part by National Natural Science Foundation of China (No. 62476052), Sichuan Science and Technology Program (No. 2024NSFSC1473), the open project of Sichuan Provincial Key Laboratory of Philosophy and Social Science for Language Intelligence in Special Education (No. YYZN-2023-3), the Humanities and Social Sciences Project of the Ministry of Education (No. 23YJA740013), and Central Guidance for Local Science and Technology Development Fund Projects (No. 2024ZYD0268)." + }, + { + "type": "title", + "bbox": [ + 0.093, + 0.284, + 0.188, + 0.3 + ], + "angle": 0, + "content": "References" + }, + { + "type": "ref_text", + "bbox": [ + 0.101, + 0.31, + 0.482, + 0.392 + ], + "angle": 0, + "content": "[1] Nikolas Adaloglou, Theocharis Chatzis, Ilias Papastratis, Andreas Stergioulas, Georgios Th Papadopoulos, Vassia Zacharopoulou, George J Xydopoulos, Klimnis Atzakas, Dimitris Papazachariou, and Petros Daras. A comprehensive study on deep learning-based methods for sign language recognition. IEEE TMM, 24:1750-1762, 2021. 1, 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.395, + 0.482, + 0.451 + ], + "angle": 0, + "content": "[2] Samuel Albanie, Gül Varol, Liliane Momeni, Triantafyllos Afouras, Joon Son Chung, Neil Fox, and Andrew Zisserman. Bsl-1k: Scaling up co-articulated sign language recognition using mouthing cues. In ECCV. Springer, 2020. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.453, + 0.482, + 0.495 + ], + "angle": 0, + "content": "[3] Gedas Bertasius, Heng Wang, and Lorenzo Torresani. Is space-time attention all you need for video understanding? In ICML, page 4, 2021. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.497, + 0.482, + 0.55 + ], + "angle": 0, + "content": "[4] Amanda Cardoso Duarte, Samuel Albanie, Xavier Giro Nieto, and Gül Varol. Sign language video retrieval with freeform textual queries. In CVPR, pages 14074-14084, 2022. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.554, + 0.482, + 0.597 + ], + "angle": 0, + "content": "[5] Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. In CVPR, pages 6299-6308, 2017. 2, 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.599, + 0.482, + 0.641 + ], + "angle": 0, + "content": "[6] Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. In CVPR, pages 6299-6308, 2017. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.643, + 0.482, + 0.697 + ], + "angle": 0, + "content": "[7] Yutong Chen, Fangyun Wei, Xiao Sun, Zhirong Wu, and Stephen Lin. A simple multi-modality transfer learning baseline for sign language translation. In CVPR, pages 5120-5130, 2022. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.7, + 0.482, + 0.756 + ], + "angle": 0, + "content": "[8] Ke Cheng, Yifan Zhang, Congqi Cao, Lei Shi, Jian Cheng, and Hanqing Lu. Decoupling gcn with dropgraph module for skeleton-based action recognition. In ECCV, pages 536-553. Springer, 2020. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.758, + 0.482, + 0.813 + ], + "angle": 0, + "content": "[9] Xiangxiang Chu, Zhi Tian, Yuqing Wang, Bo Zhang, Haibing Ren, Xiaolin Wei, Huaxia Xia, and Chunhua Shen. Twins: Revisiting the design of spatial attention in vision transformers. NeurIPS, 34:9355-9366, 2021. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.816, + 0.482, + 0.857 + ], + "angle": 0, + "content": "[10] MMPose Contributors. Openmmlab pose estimation toolbox and benchmark. https://github.com/open-mmlab/mmpose, 2020.3" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.859, + 0.482, + 0.902 + ], + "angle": 0, + "content": "[11] Jeonghyeok Do and Munchurl Kim. Skateformer: Skeletal-temporal transformer for human action recognition. arXiv preprint arXiv:2403.09508, 2024. 3, 5" + }, + { + "type": "list", + "bbox": [ + 0.094, + 0.31, + 0.482, + 0.902 + ], + "angle": 0, + "content": null + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.093, + 0.905, + 0.135 + ], + "angle": 0, + "content": "[12] Yong Du, Wei Wang, and Liang Wang. Hierarchical recurrent neural network for skeleton based action recognition. In CVPR, pages 1110-1118, 2015. 1, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.137, + 0.905, + 0.178 + ], + "angle": 0, + "content": "[13] Karen Emmorey. Language, cognition, and the brain: Insights from sign language research. Psychology Press, 2001. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.181, + 0.905, + 0.235 + ], + "angle": 0, + "content": "[14] Hezhen Hu, Weichao Zhao, Wengang Zhou, Yuechen Wang, and Houqiang Li. Signbert: Pre-training of hand-model-aware representation for sign language recognition. In ICCV, pages 11087-11096, 2021. 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.238, + 0.905, + 0.279 + ], + "angle": 0, + "content": "[15] Hezhen Hu, Wengang Zhou, and Houqiang Li. Hand-model-aware sign language recognition. In AAAI, pages 1558-1566, 2021. 2, 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.282, + 0.905, + 0.35 + ], + "angle": 0, + "content": "[16] Hezhen Hu, Wengang Zhou, Junfu Pu, and Houqiang Li. Global-local enhancement network for nmf-aware sign language recognition. ACM transactions on multimedia computing, communications, and applications (TOMM), 17(3): 1-19, 2021. 2, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.353, + 0.905, + 0.408 + ], + "angle": 0, + "content": "[17] Hezhen Hu, Weichao Zhao, Wengang Zhou, and Houqiang Li. Signbert+: Hand-model-aware self-supervised pretraining for sign language understanding. IEEE TPAMI, 45 (9):11221-11239, 2023. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.411, + 0.905, + 0.452 + ], + "angle": 0, + "content": "[18] Lianyu Hu, Liqing Gao, Zekang Liu, and Wei Feng. Dynamic spatial-temporal aggregation for skeleton-aware sign language recognition. In LREC/COLING, 2024. 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.454, + 0.905, + 0.496 + ], + "angle": 0, + "content": "[19] Jie Huang, Wengang Zhou, Houqiang Li, and Weiping Li. Attention-based 3d-cnns for large-vocabulary sign language recognition. IEEE TCSVT, 29(9):2822-2832, 2018. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.498, + 0.905, + 0.539 + ], + "angle": 0, + "content": "[20] Jie Huang, Wengang Zhou, Qilin Zhang, Houqiang Li, and Weiping Li. Video-based sign language recognition without temporal segmentation. In AAAI, 2018. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.542, + 0.905, + 0.583 + ], + "angle": 0, + "content": "[21] Songyao Jiang, Bin Sun, Lichen Wang, Yue Bai, Kunpeng Li, and Yun Fu. Skeleton aware multi-modal sign language recognition. In CVPR, pages 3413-3423, 2021. 1, 3, 4, 5, 8" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.586, + 0.905, + 0.64 + ], + "angle": 0, + "content": "[22] Songyao Jiang, Bin Sun, Lichen Wang, Yue Bai, Kunpeng Li, and Yun Fu. Sign language recognition via skeleton-aware multi-model ensemble. arXiv preprint arXiv:2110.06161, 2021. 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.643, + 0.905, + 0.685 + ], + "angle": 0, + "content": "[23] Tao Jiang, Necati Cihan Camgoz, and Richard Bowden. Skeleton: Skeletal transformers for robust body-posed estimation. In CVPR, pages 3394–3402, 2021. 1, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.687, + 0.905, + 0.742 + ], + "angle": 0, + "content": "[24] Sheng Jin, Lumin Xu, Jin Xu, Can Wang, Wentao Liu, Chen Qian, Wanli Ouyang, and Ping Luo. Whole-body human pose estimation in the wild. In ECCV, pages 196-214. Springer, 2020. 3, 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.745, + 0.905, + 0.786 + ], + "angle": 0, + "content": "[25] Trevor Johnston and Adam Schembri. Australian Sign Language (Auslan): An introduction to sign language linguistics. Cambridge University Press, 2007. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.788, + 0.905, + 0.842 + ], + "angle": 0, + "content": "[26] Hamid Reza Vaezi Joze and Oscar Koller. Ms-asl: A large-scale data set and benchmark for understanding american sign language. arXiv preprint arXiv:1812.01053, 2018. 2, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.846, + 0.905, + 0.902 + ], + "angle": 0, + "content": "[27] Alexander Kapitanov, Kvanchiani Karina, Alexander Nagaev, and Petrova Elizaveta. Slovo: Russian sign language dataset. In International Conference on Computer Vision Systems, pages 63-73. Springer, 2023. 2, 7, 8" + }, + { + "type": "list", + "bbox": [ + 0.517, + 0.093, + 0.905, + 0.902 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "24328" + } + ], + [ + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.092, + 0.482, + 0.133 + ], + "angle": 0, + "content": "[28] Oscar Koller. Quantitative survey of the state of the art in sign language recognition. arXiv preprint arXiv:2008.09918, 2020. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.134, + 0.482, + 0.189 + ], + "angle": 0, + "content": "[29] Oscar Koller, Sepehr Zargaran, Hermann Ney, and Richard Bowden. Deep sign: Enabling robust statistical continuous sign language recognition via hybrid cnn-hmms. *IJCV*, 126: 1311–1325, 2018. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.19, + 0.482, + 0.232 + ], + "angle": 0, + "content": "[30] Taeryung Lee, Yeonguk Oh, and Kyoung Mu Lee. Human part-wise 3d motion context learning for sign language recognition. In ICCV, pages 20740-20750, 2023. 1, 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.233, + 0.482, + 0.287 + ], + "angle": 0, + "content": "[31] Chao Li, Qiaoyong Zhong, Di Xie, and Shiliang Pu. Cooccurrence feature learning from skeleton data for action recognition and detection with hierarchical aggregation. arXiv preprint arXiv:1804.06055, 2018. 1, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.288, + 0.482, + 0.357 + ], + "angle": 0, + "content": "[32] Dongxu Li, Cristian Rodriguez, Xin Yu, and Hongdong Li. Word-level deep sign language recognition from video: A new large-scale dataset and methods comparison. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 1459-1469, 2020. 2, 7, 8" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.358, + 0.482, + 0.412 + ], + "angle": 0, + "content": "[33] Dongxu Li, Xin Yu, Chenchen Xu, Lars Petersson, and Hongdong Li. Transferring cross-domain knowledge for video sign language recognition. In CVPR, pages 6205-6214, 2020. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.413, + 0.482, + 0.468 + ], + "angle": 0, + "content": "[34] Maosen Li, Siheng Chen, Xu Chen, Ya Zhang, Yanfeng Wang, and Qi Tian. Actional-structural graph convolutional networks for skeleton-based action recognition. In CVPR, pages 3595–3603, 2019. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.469, + 0.482, + 0.537 + ], + "angle": 0, + "content": "[35] Yanghao Li, Chao-Yuan Wu, Haoqi Fan, Karttikeya Mangalam, Bo Xiong, Jitendra Malik, and Christoph Feichtenhofer. Mviv2: Improved multiscale vision transformers for classification and detection. In CVPR, pages 4804-4814, 2022. 7, 8" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.538, + 0.482, + 0.594 + ], + "angle": 0, + "content": "[36] Jiajia Liao, Yujun Liu, Yingchao Piao, Jinhe Su, Guorong Cai, and Yundong Wu. Gle-net: A global and local ensemble network for aerial object detection. International Journal of Computational Intelligence Systems, 15(1):2, 2022. 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.595, + 0.482, + 0.649 + ], + "angle": 0, + "content": "[37] Jinfu Liu, Xinshun Wang, Can Wang, Yuan Gao, and Mengyuan Liu. Temporal decoupling graph convolutional network for skeleton-based gesture recognition. IEEE TMM, 26:811-823, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.65, + 0.482, + 0.692 + ], + "angle": 0, + "content": "[38] Ze Liu, Jia Ning, Yue Cao, Yixuan Wei, Zheng Zhang, Stephen Lin, and Han Hu. Video swim transformer. In CVPR, pages 3202-3211, 2022. 3, 7, 8" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.692, + 0.482, + 0.734 + ], + "angle": 0, + "content": "[39] Anna Mindess. Reading between the signs: Intercultural communication for sign language interpreters. Nicholas Brealey, 2014. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.735, + 0.482, + 0.789 + ], + "angle": 0, + "content": "[40] Liliane Momeni, Hannah Bull, KR Prajwal, Samuel Albanie, Gül Varol, and Andrew Zisserman. Automatic dense annotation of large-vocabulary sign language videos. In ECCV, pages 671-690. Springer, 2022. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.79, + 0.482, + 0.844 + ], + "angle": 0, + "content": "[41] Evonne Ng, Shiry Ginosar, Trevor Darrell, and Hanbyul Joo. Body2hands: Learning to infer 3d hands from conversational gesture body dynamics. In CVPR, pages 11865-11874, 2021. 1, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.845, + 0.482, + 0.9 + ], + "angle": 0, + "content": "[42] Ladislav Rampášek, Michael Galkin, Vijay Prakash Dwivedi, Anh Tuan Luu, Guy Wolf, and Dominique Beaini. Recipe for a general, powerful, scalable graph transformer. NeurIPS, 35:14501-14515, 2022. 3" + }, + { + "type": "list", + "bbox": [ + 0.093, + 0.092, + 0.482, + 0.9 + ], + "angle": 0, + "content": null + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.092, + 0.905, + 0.134 + ], + "angle": 0, + "content": "[43] Razieh Rastgoo, Kourosh Kiani, and Sergio Escalera. Sign language recognition: A deep survey. Expert Systems with Applications, 164:113794, 2021. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.135, + 0.904, + 0.162 + ], + "angle": 0, + "content": "[44] Wendy Sandler. Sign language and linguistic universals. Cambridge University, 2006. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.163, + 0.905, + 0.204 + ], + "angle": 0, + "content": "[45] Noha Sarhan and Simone Frintrop. Unraveling a decade: a comprehensive survey on isolated sign language recognition. In ICCV, pages 3210-3219, 2023. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.205, + 0.905, + 0.258 + ], + "angle": 0, + "content": "[46] Lei Shi, Yifan Zhang, Jian Cheng, and Hanqing Lu. Two-stream adaptive graph convolutional networks for skeleton-based action recognition. In CVPR, pages 12026-12035, 2019. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.26, + 0.905, + 0.314 + ], + "angle": 0, + "content": "[47] Lei Shi, Yifan Zhang, Jian Cheng, and Hanqing Lu. Skeleton-based action recognition with multi-stream adaptive graph convolutional networks. IEEE TIP, 29:9532-9545, 2020. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.316, + 0.905, + 0.371 + ], + "angle": 0, + "content": "[48] Chenyang Si, Wentao Chen, Wei Wang, Liang Wang, and Tieniu Tan. An attention enhanced graph convolutional LSTM network for skeleton-based action recognition. In CVPR, pages 1227-1236, 2019. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.372, + 0.905, + 0.412 + ], + "angle": 0, + "content": "[49] Neil Song and Yu Xiang. Slgtformer: An attention-based approach to sign language recognition. arXiv preprint arXiv:2212.10746, 2022. 3, 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.413, + 0.905, + 0.482 + ], + "angle": 0, + "content": "[50] Yi-Fan Song, Zhang Zhang, Caifeng Shan, and Liang Wang. Stronger, faster and more explainable: A graph convolutional baseline for skeleton-based action recognition. In proceedings of the 28th ACM international conference on multimedia, pages 1625-1633, 2020. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.484, + 0.905, + 0.525 + ], + "angle": 0, + "content": "[51] Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. Deep high-resolution representation learning for human pose estimation. In CVPR, pages 5693-5703, 2019. 3, 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.525, + 0.905, + 0.553 + ], + "angle": 0, + "content": "[52] Clayton Valli. Linguistics of American Sign Language: An Introduction. Gallaudet University Press, 2000. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.554, + 0.905, + 0.607 + ], + "angle": 0, + "content": "[53] Gül Varol, Liliane Momeni, Samuel Albanie, Triantafyllos Afouras, and Andrew Zisserman. Scaling up sign spotting through sign language dictionaries. *IJCV*, 130(6):1416–1439, 2022. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.608, + 0.905, + 0.662 + ], + "angle": 0, + "content": "[54] Hongsong Wang and Liang Wang. Modeling temporal dynamics and spatial configurations of actions using two-stream recurrent neural networks. In CVPR, pages 499-508, 2017. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.664, + 0.905, + 0.718 + ], + "angle": 0, + "content": "[55] Hanjie Wang, Xiujuan Chai, and Xin Chen. A novel sign language recognition framework using hierarchical grassmann covariance matrix. IEEE TMM, 21(11):2806-2814, 2019. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.719, + 0.905, + 0.774 + ], + "angle": 0, + "content": "[56] Chengcheng Wei, Jian Zhao, Wengang Zhou, and Houqiang Li. Semantic boundary detection with reinforcement learning for continuous sign language recognition. IEEE TCSVT, 31(3):1138-1149, 2020." + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.775, + 0.905, + 0.816 + ], + "angle": 0, + "content": "[57] Pan Xie, Mengyi Zhao, and Xiaohui Hu. Pisltrc: Position-informed sign language transformer with content-aware convolution. IEEE TMM, 24:3908-3919, 2021. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.817, + 0.905, + 0.859 + ], + "angle": 0, + "content": "[58] Sijie Yan, Yuanjun Xiong, and Dahua Lin. Spatial temporal graph convolutional networks for skeleton-based action recognition. In AAAI, 2018. 1, 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.516, + 0.86, + 0.905, + 0.901 + ], + "angle": 0, + "content": "[59] Aoxiong Yin, Zhou Zhao, Weike Jin, Meng Zhang, Xingshan Zeng, and Xiaofei He. Mlslt: Towards multilingual sign language translation. In CVPR, pages 5109-5119, 2022. 1" + }, + { + "type": "list", + "bbox": [ + 0.516, + 0.092, + 0.905, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.945, + 0.52, + 0.957 + ], + "angle": 0, + "content": "24329" + } + ], + [ + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.092, + 0.482, + 0.147 + ], + "angle": 0, + "content": "[60] Weichao Zhao, Hezhen Hu, Wengang Zhou, Jiaxin Shi, and Houqiang Li. Best: Bert pre-training for sign language recognition with coupling tokenization. In AAAI, pages 3597-3605, 2023. 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.149, + 0.482, + 0.204 + ], + "angle": 0, + "content": "[61] Weichao Zhao, Hezhen Hu, Wengang Zhou, Yunyao Mao, Min Wang, and Houjiang Li. Masa: Motion-aware masked autoencoder with semantic alignment for sign language recognition. IEEE TCSVT, 2024. 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.206, + 0.482, + 0.261 + ], + "angle": 0, + "content": "[62] Weichao Zhao, Wengang Zhou, Hezhen Hu, Min Wang, and Houqiang Li. Self-supervised representation learning with spatial-temporal consistency for sign language recognition. arXiv preprint arXiv:2406.10501, 2024. 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.263, + 0.482, + 0.304 + ], + "angle": 0, + "content": "[63] Ronglai Zuo, Fangyun Wei, and Brian Mak. Natural language-assisted sign language recognition. In CVPR, pages 14890-14900, 2023. 6, 7" + }, + { + "type": "list", + "bbox": [ + 0.093, + 0.092, + 0.482, + 0.304 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.946, + 0.519, + 0.956 + ], + "angle": 0, + "content": "24330" + } + ] +] \ No newline at end of file diff --git a/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/1135c1a4-3af0-45ac-a497-2c9557104d8c_origin.pdf b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/1135c1a4-3af0-45ac-a497-2c9557104d8c_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..da66e0c304d12275c33555275c4bab461a356461 --- /dev/null +++ b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/1135c1a4-3af0-45ac-a497-2c9557104d8c_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25ea6623303544c6406f405e9b1bdbebd55a0a5bf1bd89647bb6cfea9ddc822c +size 3520878 diff --git a/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/full.md b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/full.md new file mode 100644 index 0000000000000000000000000000000000000000..7f08fd4b4777e70abe06f19a8d3ad6d7f9c1167b --- /dev/null +++ b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/full.md @@ -0,0 +1,331 @@ +# VSNet: Focusing on the Linguistic Characteristics of Sign Language + +Yuhao Li $^{1}$ , Xinyue Chen $^{1}$ , Hongkai Li $^{1}$ , Xiaorong Pu $^{1,3}$ , Peng Jin $^{2}$ , Yazhou Ren $^{1,3,*}$ $^{1}$ School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China; $^{2}$ Sichuan Provincial Key Lab of Philosophy and Social Science for Language Intelligence in Special Education, Leshan Normal University, Leshan, China; $^{3}$ Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China, Shenzhen, China + +# Abstract + +Sign language is a visual language expressed through complex movements of the upper body. The human skeleton plays a critical role in sign language recognition due to its good separation from the video background. However, mainstream skeleton-based sign language recognition models often overly focus on the natural connections between joints, treating sign language as ordinary human movements, which neglects its linguistic characteristics. We believe that just as letters form words, each sign language gloss can also be decomposed into smaller visual symbols. To fully harness the potential of skeleton data, this paper proposes a novel joint fusion strategy and a visual symbol attention model. Specifically, we first input the complete set of skeletal joints, and after dynamically exchanging joint information, we discard the parts with the weakest connections to other joints, resulting in a fused, simplified skeleton. Then, we group the joints most likely to express the same visual symbol and discuss the joint movements within each group separately. To validate the superiority of our method, we conduct extensive experiments on multiple public benchmark datasets. The results show that, without complex pretraining, we still achieve new state-of-the-art performance. The code is available at https://github.com/atinyboy/VSNet. + +# 1. Introduction + +Sign language is a visual language expressed through body movements and serves as the primary communication tool for the deaf community [13]. Although sign language and spoken language have distinctly different grammatical structures [1, 7], they share common linguistic features [39, 44, 59]. Accurate sign language translation can facilitate communication between deaf and hearing people, fostering a more inclusive and accessible society. Among + +![](images/0ef7164ed9cc0e33de67ef10e8993b7a4a21ae4ff844e5a166b44fb3d111af05.jpg) + +![](images/44fe15157fcbe074d6c1d7cd80d06ce1f7f18531872187c473e9164ff0c00f43.jpg) +gloss: 'accent' + +![](images/f9bf732e363bf30348c2d295f318f5d05c852e71aa7b797239453d99c7a09b96.jpg) + +![](images/ca5a47bc0d09f7265a256a59d8607956570ca3711d2ec5e38228d7f2796734ad.jpg) + +![](images/fdeac63056da7ab56e05b49bb7c6ccfa708b60992d811714809658431cc4ca36.jpg) +Figure 1. In the sign for "accent", the index and middle fingers (or only index finger) are extended toward the neck while other fingers are curled, forming the "pointing" VS. In the sign for "algebra", movements can be decomposed into multiple VS such as "crossed arms," "fist," and "thumb-up." + +![](images/18fa9e3dbae1a5067929bcacce7062c5be4971a7cddd1ad711e479f11e05de4b.jpg) +gloss:'algebra' + +![](images/ad3e18e00f9ec6d67c41252c0742477cc98782276705a52e360e368fa82d5f19.jpg) + +![](images/10785b8de7ae97caba6d78a7d9b114606070865a8f7c70ff8eea76801732b85a.jpg) + +the key subtasks in sign language translation is isolated sign language recognition (ISLR), which focuses on translating individual glosses. ISLR has garnered significant attention as it aids people in accurately understanding numerous gestures and quickly learning sign language. + +In recent years, human skeletons have been increasingly applied in the ISLR field due to their strong operability and clear separation from video backgrounds. Although most skeleton-based recognition methods have achieved promising results, they typically treat sign language as typical human motion [12, 21, 23, 30, 31, 41, 58], overlooking the intrinsic relationships among joints and the unique linguistic characteristics [25, 52] of sign language itself. We believe that: (1) The importance of joints in sign language is not equivalent; joints with stronger connections to others are more significant than isolated joints, and connections among joints with similar movement tendencies hold more meaning than those among dispersed joints. (2) Similar to how letters form words, each sign language gloss can be decomposed into smaller visual symbols (VS). While VS + +lack explicit meaning on their own, from a skeletal perspective, they capture distinct movement patterns of joints, effectively decomposing sign language actions (see Fig. 1). Therefore, our work focuses on these two core aspects: + +Core 1: Focus on identifying essential joints. Including too many joints can impede training, while too few can lead to information loss. On the one hand, many prior models have used manually trimmed, simplified skeletons, but manually selecting the optimal joints is not trivial. On the other hand, enabling a model to automatically choose key joints can be computationally intensive. Fortunately, the dynamically learned skeletons from GCNs approximate real structures more closely, enabling us to reduce the dependency on an initial skeleton. By computing the weight of each joint, we can identify and retain the most important joints while discarding less critical local joints. Specifically, for the complete hand skeleton, we discard a small number of joints based on computed weights after each GCN layer, relearning new skeleton connections in the subsequent GCN layer. We control the number of joints discarded within each region to ensure that the final GCN output retains a recognizable structure. Notably, even joints with the lowest weights undergo at least one convolution, so their information is not entirely discarded. + +Core 2: Focus on emphasizing visual symbols (VS) in sign language. If each joint is considered the smallest element of a sign language action, then the combination of joints across all frames becomes too vast to compute self-attention. In practice, sign language is not random motion; it is designed for human recognition. When a signer interprets sign language, they do not scrutinize the position of every joint but instead form a general impression of each hand. Thus, in this work, we concentrate on recognizing VS. Unfortunately, isolating each VS from movements is as challenging as the entire SLR task itself. Therefore, we adopt a simplified approach, segmenting the skeleton into a generalized VS in a fixed manner, as shown in Fig. 2. We then design two self-attention modules to capture spatial and temporal relationships between these VS. Empirically, we find that this approach significantly enhances the model's discriminative power. + +Our main contributions are summarized as follows: + +- We propose a novel module with an automated skeleton simplification function for sign language recognition. This module is designed to dynamically learn global skeleton connections while progressively discarding isolated local joints. +- Considering the linguistic characteristics of sign language, we introduce a new self-attention model called VSformer, which captures temporal and spatial patterns in sign language motions and extracts knowledge from the human skeleton based on the concept of VS. +- We conduct extensive experiments on ISLR to validate + +![](images/643f5b9a8f8824d4be18635ddea1b3f7e73bb10c6cc91a7c3c93649d5a9787aa.jpg) +Figure 2. Decomposing generalized visual symbols from videos + +the effectiveness of our proposed approach. Compared to previous methods, our approach achieves SoTA performance on four public benchmarks, namely WLASL [32], MSASL [26], NMFs-CSR [16], and Slovo [27]. + +# 2. Related Works + +# 2.1. Sign Language Recognition + +Sign language is the primary means of communication within the deaf community. With significant advances in machine learning and computer vision over the past decade, exploring sign language recognition (SLR) has become increasingly important [28, 43, 45], as it can automatically interpret sign language and help deaf and mute individuals communicate more easily in daily life. Unlike directly translating sign language sentences, isolated sign language recognition (ISLR) is a subtask in the field of sign language translation [45], aimed at translating each individual gloss [40, 53]. ISLR is not only useful for guiding beginners in learning sign language and assisting in the creation of new sign language gestures, but it also lays a solid foundation for SLR [1, 4, 53]. Research in this area can be divided into two categories based on the input modality: RGB-based methods and skeleton-based methods. + +RGB-based Recognition. RGB-based methods, similar to other video recognition tasks, typically use CNNs to extract visual features from RGB videos [5, 15, 19, 20, 29, 55-57], and then aggregate temporal and spatial information to predict sign language. However, RGB-based models face two significant challenges: first, they are heavily affected by changes in the video background, and second, video data occupies a large amount of memory resources. + +Skeleton-based Recognition. Compared to RGB images, human skeletons [8, 47, 50, 54] describe the motion information conveyed in sign language videos using simplified points and edges, reducing interference from complex backgrounds and the appearance of the signer. Moreover, operations on three-dimensional spatial points with actual physical meanings offer stronger interpretability and + +![](images/a0fe65ed0df5a97890f903093b93c51daf44c0b86f7de44230e2f011d4ddd927.jpg) +Figure 3. Overview of our VSNet. Here, Down sample consists of a convolution layer with a kernel size of 7 and a stride of 2 along the temporal dimension, followed by a batch norm layer, Average Pooling represent transforming the data into a single-dimensional feature by taking the mean across both temporal and spatial dimensions. For details on the Skeleton- $(a,b,c)$ , please refer to Fig. 4. + +lower memory consumption. In earlier works, identifying the correct human keypoints in a video was a major challenge. However, with the development of deep learning, human pose estimation models have become increasingly reliable [10, 24, 51]. + +# 2.2. Human Skeleton + +Due to the natural physical connections between skeleton joints, Graph Convolutional Network (GCN) has been widely applied to model spatio-temporal relationships [12, 23, 31, 34, 41, 46]. GCN require predefined connectivity of the key points to construct the graph network. However, the physically natural skeleton connections may not always be the most suitable edges for recognition. Relying solely on the skeleton itself to construct the graph can limit the information flow between nodes. To address this issue, SL-GCN [21] employs a dynamic graph with decoupling between different frames, while TD-GCN [37] adds supernodes to control the connections between nodes. A common drawback of these approaches is that they require complex networks to model the key points. Furthermore, while dynamic graphs alleviate the limitations of the original skeleton, blindly allowing all key points to connect introduces redundant information, leading to suboptimal models. + +Therefore, we argue that relying solely on GCN does not fully exploit the true potential of skeleton data. + +In recent years, attention-based methods [3, 6, 9, 38, 42, 48] have also made progress in human skeleton recognition. Unlike GCN, attention mechanisms do not require the researcher to provide information on joint connections, instead equally attending to all nodes. However, captur + +ing the correlations between all joints across all frames requires substantial memory resources. To address this, SLGTformer [49] decomposes the spatio-temporal pose sequence into spatial graphs and temporal windows, while SkateFormer [11] partitions joints and frames and performs self-attention on each partition. Based on these theories, we design a model where GCNs and self-attention networks complement each other. GCNs are responsible for exploring the spatial correlations between human joints, while self-attention networks focus on capturing the motion of the skeleton across multiple time sequences and spatial combinations. + +# 3. Methods + +# 3.1. Data Pre-Processing + +Key Point Extraction. Sign language is a visual language primarily conveyed through hand and body movements, with facial expressions also playing a crucial role. However, the human skeleton mainly describes bones and joints, excluding muscles, making it unable to accurately capture facial expressions in sign language videos. Our investigation revealed that a significant number of signers, especially beginners, do not exhibit extensive facial expressions during sign language communication. Therefore, our main experiments exclude facial expressions, while the appendix provides additional details on how the concept of VS can be extended to the face. + +In previous GCN models, finger joints were often simplified to avoid making the graph too complex, leading to a loss of information. In this study, we selected a complete + +![](images/3b302e9367d27a69b446d6fd883fe21af9ef8d9b70365c49823d0a11074b393e.jpg) +Figure 4. Illustration of the skeleton and grouping type-1. $V_{0}$ represents a localization node, the dashed box indicates nodes that are grouped together and hollow dots indicate dropped nodes. Please note that (a) is the input real skeleton diagram, while (b) and (c) are simplified representations drawn after dropping for ease of analysis. + +![](images/b715aea289b6efb25c9f8e0f180402f2a4f181c1c5aefe14eb1470b89cdc6b34.jpg) +Figure 5. Skeleton groupings type-2 and type-3. The other hand is grouped in a mirrored fashion. + +set of 44 key points from both hands out of the 133 human body joints extracted using HRNet [24, 51], as shown in Fig. 4(a). + +Data Enhancing. Given a video $X \in \mathbb{R}^{T \times V \times C}$ , where T, V, and C represent the number of frames, joints, and coordinates respectively. We replace abnormal points—such as those out of range (less than 0, more than 512) or clearly distorted—caused by model extraction errors, with the average value from adjacent frames. Additionally, to standardize body positions and movement amplitudes across different signers in the sign language videos, we first use the hollow node $V_{0}$ in Fig. 4 as the reference node for the entire video. Then, all key points in each frame are shifted based on this reference coordinate, as follows: + +$$ +X (t, v, c) = X (t, v, c) - X (t, v _ {0}, c) + [ x, y, z ] \tag {1} +$$ + +where $[x,y,z]$ is the average value of all the coordinates of $V_{0}$ in the dataset. + +In the time dimension, we randomly select 64 frames to standardize the video length and perform random mirroring and shifting of the video to generate augmented data. + +# 3.2. Self-Pacing Dropping Block (SPD) + +As shown in Fig. 3, our SPD consists of two components: "Keypoint Activation" and "Weak Joints Dropping". The input is a complete hand skeleton where the node information is independent, and the output is a high-dimensional feature skeleton after activation. The entire block is stacked three times, where the third block only outputs features and does not build a new skeleton. + +Keypoint Activation Layer. In this part, we define the interaction of each joint with others as an activation. We employ a decoupled graph convolution [21] to perform this activation, represented as: + +$$ +F _ {o u t} ^ {t} = \frac {1}{p} \sum_ {k = 0} ^ {p} D ^ {- \frac {1}{2}} A _ {k} D ^ {- \frac {1}{2}} \phi \left(F _ {i n} ^ {t}\right) \tag {2} +$$ + +where $A$ is the learnable adjacency matrix initialized as $I + In + Out$ , $D$ presents the diagonal degree of $A$ . Here, $I$ denotes self-connections, while $In$ and $Out$ represent inward and outward connections, respectively. The symbol $\phi$ represents a linear layer, and $p$ represents the decoupling quantity, with a different $A$ applied each time, followed by averaging the results. Following the graph convolution, we apply the simplest Temporal Convolutional Network (TCN) to capture features along the temporal dimension. + +Weak Joints Dropping Layer. Compared to the simplified skeletons used in [21, 49], our approach utilizes a complete set of skeleton keypoints that contains richer information. However, this comes at the cost of significantly increasing the model's burden and raising the risk of introducing outlier points. To leverage these additional points with minimal overhead, we have designed an adaptive skeleton simplification strategy. + +Before formally introducing this layer, it is necessary to present our approach to skeleton decomposition. Based on the definition of VS described in Introduction, the decomposition concept involves grouping joints that exhibit strong overall correlation or frequently form VS together in sign language. These groupings can be viewed as a generalized or macro-level VS. A simple approach is grouping each finger and arm separately, but this creates too many groups, potentially affecting later interactions. Based on this, we conducted various experiments on VS, including treating individual fingers and distant fingers as VS. However, these experiments did not yield optimal results (the average performance in WLASL dataset decreased by $1.20\%$ and $1.64\%$ , respectively). We believe this is because the information from a single finger is insufficient to form a meaningful VS, while the information disparity between distant fingers is even greater. This also serves as the skeleton grouping basis for Figs. 4 and 5. + +After graph convolution in Eq. 2, $A(i,j)$ has learned the weighted edge between node $i$ and node $j$ . These edges reflect the connection strength between $i$ and $j$ . Since these + +connections are not normalized via softmax, we can use $A$ as a guide to calculate the total correlation of all keypoints, Cor, as follows: + +$$ +C o r _ {i} = \sum_ {j} \max \left(A _ {0} (i, j), A _ {1}, \dots , A _ {p}\right) \tag {3} +$$ + +A lower $Cor$ to other keypoints indicates that the point is more independent. We refer to the two most independent points in each group as weak joints, which are dropped from the skeleton. To ensure that the number of keypoints in each group remains consistent, the groups involving both arms will have two fewer joints dropped. + +# 3.3.VSformer + +After the joint fusion block, we obtained six different joint groupings, as shown in Fig. 4(c). Based on the concept of skeletal disassembly, we treat each grouping as a generalized VS. We now need to focus on two main questions: Q1, what are the relationships between different VS? And Q2, do these VS remain consistent throughout the entire video, or do they change after a brief period? + +Inspired by [11], which proposes an efficient model for partitioning human limbs in action recognition, we employ two types of self-attention to constrain the model's focus on the aforementioned questions. Specifically, we calculate spatial pairing attention among different combinations within a frame to address Q1, and we compute temporal window attention for both global and local motion of the combinations across different frames to address Q2. + +Spatial Pairing Attention (SPA). Given a frame $F_{t} \in \mathbb{R}^{G \times C}$ , where $G = 6$ represents VS groupings, the ideal approach would involve calculating attention among all VS pairwise. However, this is inefficient and often purposeless, as VS that are far apart in the skeletal structure tend to have weak correlations. For instance, in the sign language gesture for "accent", it is unnecessary to compare the fingers of the left hand with those of the unused right hand. Based on this observation, we ultimately partition $F_{t}$ into pairs as follows: + +$$ +F _ {t} = \left[ P _ {\text {a r m}}, P _ {\text {l e f t}}, P _ {\text {r i g h t}} \right] \tag {4} +$$ + +where $P_{\mathrm{arm}} \in \mathbb{R}^{2 \times C}$ represents the grouping of both arms, $P_{\mathrm{left}}$ denotes the four fingers of the left hand, and $P_{\mathrm{right}}$ represents the remaining four fingers of the right hand. Represented as: + +$$ +S P A _ {\text {o u t}} = \operatorname {c o n c a t} \left(\operatorname {s o f t m a x} \left(\frac {Q _ {i} K _ {i} ^ {T}}{\sqrt {h}}\right) V _ {i}\right) \tag {5} +$$ + +where $Q_{i}, K_{i}, V_{i}$ are obtained from $P_{i}$ after mapping through linear layers. The attention is actually calculated among the 8 joints within each pairing and is ultimately concatenated to form the output. + +Temporal Window Attention (TWA). Depending on different sign language videos, the VS can be classified into + +![](images/621111fb7906cca097896dddac1e940b4f52be4a0b9723b1c4add3959b904881.jpg) +Figure 6. TWA windows. The small ticks represent frames, and the large ticks represent windows. + +two types: short duration and long duration, representing their respective duration periods. Accordingly, our TWA is also divided into two types: TWA-s and TWA-l. Given a video $X \in \mathbb{R}^{T \times G \times C}$ , where $T$ represents the total number of frames, the timeline is divided into $n$ windows, each containing $k$ frames, such that $T = n \times k$ , as shown in Fig. 6. For TWA-s, attention is computed within each window. For TWA-l, frames with the same index are selected sequentially from each window. Finally, the results from both types of attention are concatenated to produce the output, represented as: + +$$ +T W A _ {o u t} = \operatorname {s o f t m a x} \left(\phi_ {1} (W) \phi_ {2} ^ {T} (W) + b\right) \phi_ {3} (W) \tag {6} +$$ + +where $W$ represents the selected windows, $\phi_{n}$ denotes dividing the $C$ dimension into three equal parts and selecting the $n$ -th part, and $b$ represents the added bias. + +It is worth noting that the discussions on Q1 and Q2 are not fully addressed by these two attention layers. The SPA only addresses the relationships between paired VS, while the windows defined in TWA do not entirely correspond to the duration of the VS. Therefore, we incorporate a simple GCN and TCN to enhance the model's ability to address the remaining scenarios. The connection method of SPA and TWA is shown in the right side of Fig. 3. After passing the input data $X \in \mathbb{R}^{T \times G \times C}$ through a linear layer, we divide its channel dimension $C$ into four equal parts, sending each part into one of the four modules and later concatenated. Additionally, unlike the Decoupled GCN in Sect. 3.2, the GCN here utilizes an adjacency matrix shared across certain frames, which significantly reduces the computational load. + +# 3.4. Multi-Grouping Ensemble + +In the previous sections, we introduced the concept of skeletal disassembly based on VS. However, since we cannot precisely determine the true VS composition for each sign language, using a fixed disassembly method appears to be a compromise, potentially weakening interpretability. Therefore, we adopt an ensemble strategy to reduce such discrepancies. Specifically, we designed two additional skeleton grouping schemes, as shown in the Fig. 5, and trained the model separately to obtain the output $q$ from the final fully connected layer for each scheme. Following [21], we aggregate these results using a weighted sum, represented as: $q_{\mathrm{total}} = \alpha_1q_{\mathrm{type - 1}} + \alpha_2q_{\mathrm{type - 2}} + \alpha_3q_{\mathrm{type - 3}} + \alpha_4q_{\mathrm{bone}}$ , where $\alpha$ represents the weights. For a detailed discussion on $\alpha$ , + +
MethodsMSASL1000MSASL200MSASL100
P-IP-CP-IP-CP-IP-C
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
RGB-based
HMA [15]49.1669.7546.2768.6066.3084.0367.4784.0373.4589.7074.5989.70
I3D [5]*--57.6981.05--81.9793.79--81.7695.16
BSL [2]64.7185.5961.5584.43--------
TCK [33]†----80.3191.8281.1492.2483.0493.4683.9193.52
skeleton-based
ST-GCN [58]36.0359.9232.3257.1552.9176.6754.2077.6259.8482.0360.7982.96
BEST [60]58.8281.1854.8780.0576.6091.5476.7591.9580.9895.1181.2495.44
SignBERT [14]†59.8081.8657.0680.9477.3491.1078.0291.4881.3793.6682.3193.76
SignBERT+ [17]†62.4283.4960.1582.4478.51-79.35-84.94-85.23-
MASA [61]†63.4783.8960.7983.2979.2592.8679.7093.3383.2295.2483.1995.46
SSRL [62]†65.2285.0962.6884.3882.9394.7883.4395.0386.2696.9686.6396.79
Ours70.7186.3668.3885.7184.6294.0485.1994.0886.2695.2486.6895.39
Ours(4-crops)73.1888.6470.8887.5986.1795.5186.6495.6188.1197.0988.3297.18
+ +Table 1. Comparison with previous work on MSASL. “†” indicates that the model was pre-trained, and “*” indicates methods that applied result fusion with other modalities in post-processing. Bold indicates the best performance, while underlined denotes the second-best. + +
MethodsWLASL2000WLASL300WLASL100
P-IP-CP-IP-CP-IP-C
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
RGB-based
I3D [5]*32.4857.31--56.1479.9456.2478.3865.8984.1167.0184.58
HMA [15]37.9171.2635.9070.00--------
BSL [2]46.8279.3644.7278.47--------
TCK [33]†----68.5689.5268.7589.4177.5291.0877.5591.42
NLA-SLR V64 [63]*51.1583.4348.1482.20--------
skeleton-based
P3D [30]44.4779.6942.1878.5267.1889.0167.6289.2476.7191.9778.2792.97
BEST [60]46.2579.3343.5277.6567.6689.2268.3189.5777.9191.4777.8392.50
SignBERT [14]†47.4683.3245.1782.3270.3688.9271.1789.3679.0793.8080.0594.17
MASA [61]†49.0682.9046.9181.8073.6591.7774.3392.1383.7293.8084.4794.30
NLA-SLR K64 [63]*49.1082.0046.1880.71--------
SL-GCN [22]*51.5084.9448.8784.02--------
DSTA [18]53.68-51.17-79.97-80.56-82.38-83.09-
Ours53.5483.1851.1882.0379.0494.0179.7494.2584.5094.5785.2594.33
Ours(4-crops)55.9887.0753.5486.0480.0994.7680.8594.9285.6694.9686.2595.08
+ +Table 2. Comparison with previous work on WLASL. “†” indicates that the model was pre-trained, and “*” indicates methods that applied result fusion with other modalities in post-processing. Bold indicates the best performance, while underlined denotes the second-best. + +please refer to Sect. 4.4. The term bone refers to the joint data of type-1 represented in vector form, generated by directing from the source joint to its target joint according to the natural connections of the human body. + +# 4. Experiments + +# 4.1. Parameter Settings + +We employ the AdamW optimizer with a plateau learning rate schedule. The initial learning rate is set to $1 \times 10^{-3}$ , with a minimum learning rate of $1 \times 10^{-5}$ and a decay rate + +of 0.5. The batch size is 32 and the loss function used is cross-entropy with label smoothing. Training spans a maximum of 200 epochs, with the first 15 epochs serving as a warm-up period starting from a learning rate of $1 \times 10^{-7}$ . A random sample of 64 frames is selected as input, and with a $50\%$ probability, image mirroring is applied. A random jitter in the range $(-10, 10)$ is added to each joint point. In the last two layers of SPD, joints are randomly masked with a probability of $(1 - p_{kr}) / 100 \times \text{epoch}$ , where $p_{kr}$ is set to 0.9. All experiments were performed with NVIDIA A100. + +
MethodsNMFs-CSR
Top-1Top-5
RGB-based
I3D [5]*64.488.0
HMA [15]64.791.0
GLE-Net [36]69.088.1
skeleton-based
ST-GCN [58]59.986.8
BEST [60]68.594.4
MASA [61]†71.797.0
SignBERT [14]†74.993.2
Ours75.395.4
Ours(4-crops)76.797.2
+ +Table 3. Comparison on NMFs-CSR. Bold indicates the best performance, while underlined denotes the second-best. + +
MethodsSlovo
Top-1Top-5
Swin-large-48 [38]55.66-
MViTv2-small-48 [35]62.18-
MViTv2-small-32 [35]64.09-
Ours76.5393.08
Ours(4-crops)77.8395.17
+ +Table 4. Comparison with the baselines on Slovo. The final number indicates the number of frames input to the model. + +Metrics. Following [22, 61, 63], we report top-1 and top-5 classification accuracy, as well as per-instance (P-I) and per-class (P-C) precision metrics. Each table includes results for single-crop and 4-crop inference, with all ablation studies conducted in the single-crop setting. For 4-crops inference, we sum the predictions from the three types of skeleton groupings along with the bone data to obtain the final prediction. + +# 4.2. Dataset + +We conduct experiments on four publicly available benchmarks: WLASL, MSASL, NMFs-CSL, and Slovo. All training occurs on the training set, with hyperparameter tuning performed on the validation set. + +MSASL [26] is a widely used dataset for American Sign Language (ASL). It contains a total of 16,054, 5,287, and 4,172 samples in the training, validation, and test sets, respectively, with a vocabulary of 1,000 signs. Two subsets, consisting of the top 100 and 300 words, are defined as MSASL100 and MSASL300, respectively. + +WLASL [32] has a larger vocabulary but fewer total samples than MSASL. It contains 2,000 signs performed by over 100 signers, with 14,289, 3,916, and 2,878 samples in the training, validation, and test sets. Similarly, two subsets, WLASL100 and WLASL300, are created. Both + +ASL datasets are collected from real-world scenarios, offering greater authenticity. + +NMFs-CSR [16] is a challenging Chinese Sign Language (CSL) dataset with a vocabulary of 1,067 and consists of 25,608 and 6,402 samples in the training and test sets, respectively. + +Slovo [27] is the latest dataset for Russian Sign Language (RSL). It contains 20,000 videos across 1,000 isolated RSL gestures from 194 signers. Notably, the authors provide 42 hand landmarks based on MediaPipe, so we do not re-extract joint points. + +# 4.3. Comparison with State-of-the-art Methods + +In this section, we compare our model with several state-of-the-art approaches. For single-modality methods, we categorize them as either RGB-based or skeleton-based. For multi-modality methods, we report the performance of their single-modality models before ensemble. + +MSASL [26]. As shown in the Tab. 1, RGB-based methods have developed slowly in recent years, and our approach shows a significant improvement when compared to them. Among skeleton-based methods, SignBERT [14] and BEST [60] designed self-supervised learning strategies to pre-train the model in order to alleviate issues related to limited interpretability and overfitting due to the small size of the SLR datasets. NASA [61] and SSRL [62] address the problem of neglecting explicit motion information and lexical signs during pre-training. However, none of these methods take into account the linguistic characteristics of sign language itself. Compared to them, our method achieves a substantial top-1 improvement of $7.96\%$ in P-I and $8.2\%$ in P-C on the full dataset. However, SSRL performs slightly better than our single-crop method on the 100-class subset, which we speculate is due to SPD requiring a certain amount of data to effectively learn the droppable nodes. + +WLASL [32]. SAM-SLR [22] and NLA-SLR [63] are currently the two best-performing heavy multi-modality ensemble models. Among them, SL-GCN is the skeleton model used in SAM-SLR for later ensemble, employing a multi-stream approach to further improve performance. K64 and V64 are skeleton and RGB models in NLASLR that use 64-frame input, where the skeleton is represented as a heatmap sequence, allowing their keypoint encoder to share the same architecture with the video encoder. DSTA [18] was the previously most accurate model. Compared to their results, our top-1 accuracy achieved a $2.30\%$ improvement on WLASL2000. + +NMFs-CSL [16]. As shown in Tab. 3, NMFs-CSL is a balanced dataset in terms of class distribution, so we only report the P-I accuracy. The pretraining strategy of SignBERT [14] is highly effective on this dataset; however, its performance on top-5 is suboptimal. Our model does not use any additional training data, yet it shows a significant + +
SkeletonMSASL1000MSASL200MSASL100WLASL2000WLASL300WLASL100NMFs-CLR
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
type-170.7186.3684.6294.0486.2695.2453.5483.1879.0494.0184.5094.9675.2695.47
bone68.4185.8882.4193.3884.6894.0650.7682.4574.7091.1784.1193.8073.6394.31
type-270.3586.1085.2894.1187.4595.6453.0283.3278.4493.4183.7293.8074.7094.86
type-370.3586.1084.9294.1185.6095.1152.2682.4978.1493.4184.8893.4174.8795.55
+ +Table 5. Performance of multi-grouping VSformer on three datasets. + +
weightingWLASL2000MSASL1000NMFs-CSL
Top-1Top-5Top-1Top-5Top-1Top-5
sum55.9887.0773.1888.6476.6597.17
acc56.0187.1873.1388.6476.6597.17
+ +Table 6. Performance of the two ensemble methods + +
MethodsWLASL300WLASL2000
Top-1Top-5Top-1Top-5
Linear74.7092.8146.3277.24
GCN76.3591.7751.1880.82
SPD(ours)79.0494.0153.5483.18
+ +Table 7. Comparing our SPD with GCN and linear layers. + +
SPATMAWLASL300WLASL2000
Top-1Top-5Top-1Top-5
✓77.6993.4152.4382.24
✓77.9993.2652.3681.83
✓✓79.0494.0153.5483.18
+ +Table 8. Ablation study on the impact of our two types of attention mechanisms. + +improvement over SignBERT in overall performance, with a $1.8\%$ improvement in top-1 accuracy and a $4\%$ improvement in top-5 accuracy. + +Slovo [27]. Slovo is a recently released RSL dataset, with only limited experimental data available. We provide an up-to-date benchmark for better evaluating this dataset. As shown in Tab. 4, the results for Swin [38] and MViTv2 [35] were provided as baselines by the dataset authors. Notably, the skeleton landmarks provided by the authors includes only 42 hand joints, without arm data, which partially limits our model's performance. Despite this, we still achieve results significantly above the baselines. + +# 4.4. Ablation Study + +In this section, we conduct several key ablation experiments on the WLASL dataset [32] and report P-I accuracy to validate the effectiveness of our approach. + +SPD. We first replace SPD with two alternative methods. As shown in Tab. 7, "Linear" denotes a three-layer Conv-2D module with the same output dimension as SPD, while "GCN" refers to three decoupled graph convolution + +layer without the weak joints dropping strategy. To ensure consistent output data shape, these two methods directly input the simplified 24 joint points. The complete SFB implementation achieves the highest accuracy, indicating that our dropping strategy better facilitates the extraction of sign language action features. + +Multi-Grouping. As shown in Tab. 5, we present the final performance of four types of crops. The overall performance of each grouping type varies only slightly, mainly depending on the sample distribution of the dataset. However, integrating all four crops significantly enhances action classification performance. The performance of the bone-based grouping is slightly lower than that of joint-based results, primarily because our model is designed for points and is less sensitive to vector-based data. + +SAM-SLR [21] ultimately using manually set weights to control the ensemble. However, we argue that such weights lack interpretability. As shown in Tab. 6, we employed two methods to integrate the results: "acc" refers to using the individual top-1 accuracy of each crop as the weight, while "sum" represents directly summing the outputs of the final layer of the model. Both methods resulted in an about $2\%$ improvement in top-1 accuracy, demonstrating that our ensemble approach is both general and robust. + +VSformer. For our two types of self-attention, we replace one with a linear layer to test the performance of the other individually. As shown in Tab. 8, the full VSformer achieves significantly better performance, indicating that features in both temporal and spatial dimensions are crucial for distinguishing sign language actions. + +# 5. Conclusion + +In this work, we design a skeleton recognition model tailored specifically for ISLR, termed VSNet. The core idea of our approach is to focus on the truly essential elements of sign language. To this end, we first introduce a weak joints dropping strategy to prioritize more critical points in the skeleton. We then propose the VSformer to capture the linguistic characteristics of sign language, where Visual Symbols (VS) represent smaller components of sign language actions. We conduct extensive experiments to validate the effectiveness of the proposed VSNet, achieving state-of-the-art performance across four benchmarks. + +# Acknowledgements + +This work was supported in part by National Natural Science Foundation of China (No. 62476052), Sichuan Science and Technology Program (No. 2024NSFSC1473), the open project of Sichuan Provincial Key Laboratory of Philosophy and Social Science for Language Intelligence in Special Education (No. YYZN-2023-3), the Humanities and Social Sciences Project of the Ministry of Education (No. 23YJA740013), and Central Guidance for Local Science and Technology Development Fund Projects (No. 2024ZYD0268). + +# References + +[1] Nikolas Adaloglou, Theocharis Chatzis, Ilias Papastratis, Andreas Stergioulas, Georgios Th Papadopoulos, Vassia Zacharopoulou, George J Xydopoulos, Klimnis Atzakas, Dimitris Papazachariou, and Petros Daras. A comprehensive study on deep learning-based methods for sign language recognition. IEEE TMM, 24:1750-1762, 2021. 1, 2 +[2] Samuel Albanie, Gül Varol, Liliane Momeni, Triantafyllos Afouras, Joon Son Chung, Neil Fox, and Andrew Zisserman. Bsl-1k: Scaling up co-articulated sign language recognition using mouthing cues. In ECCV. Springer, 2020. 6 +[3] Gedas Bertasius, Heng Wang, and Lorenzo Torresani. Is space-time attention all you need for video understanding? In ICML, page 4, 2021. 3 +[4] Amanda Cardoso Duarte, Samuel Albanie, Xavier Giro Nieto, and Gül Varol. Sign language video retrieval with freeform textual queries. In CVPR, pages 14074-14084, 2022. 2 +[5] Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. In CVPR, pages 6299-6308, 2017. 2, 6, 7 +[6] Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. In CVPR, pages 6299-6308, 2017. 3 +[7] Yutong Chen, Fangyun Wei, Xiao Sun, Zhirong Wu, and Stephen Lin. A simple multi-modality transfer learning baseline for sign language translation. In CVPR, pages 5120-5130, 2022. 1 +[8] Ke Cheng, Yifan Zhang, Congqi Cao, Lei Shi, Jian Cheng, and Hanqing Lu. Decoupling gcn with dropgraph module for skeleton-based action recognition. In ECCV, pages 536-553. Springer, 2020. 2 +[9] Xiangxiang Chu, Zhi Tian, Yuqing Wang, Bo Zhang, Haibing Ren, Xiaolin Wei, Huaxia Xia, and Chunhua Shen. Twins: Revisiting the design of spatial attention in vision transformers. NeurIPS, 34:9355-9366, 2021. 3 +[10] MMPose Contributors. Openmmlab pose estimation toolbox and benchmark. https://github.com/open-mmlab/mmpose, 2020.3 +[11] Jeonghyeok Do and Munchurl Kim. Skateformer: Skeletal-temporal transformer for human action recognition. arXiv preprint arXiv:2403.09508, 2024. 3, 5 + +[12] Yong Du, Wei Wang, and Liang Wang. Hierarchical recurrent neural network for skeleton based action recognition. In CVPR, pages 1110-1118, 2015. 1, 3 +[13] Karen Emmorey. Language, cognition, and the brain: Insights from sign language research. Psychology Press, 2001. 1 +[14] Hezhen Hu, Weichao Zhao, Wengang Zhou, Yuechen Wang, and Houqiang Li. Signbert: Pre-training of hand-model-aware representation for sign language recognition. In ICCV, pages 11087-11096, 2021. 6, 7 +[15] Hezhen Hu, Wengang Zhou, and Houqiang Li. Hand-model-aware sign language recognition. In AAAI, pages 1558-1566, 2021. 2, 6, 7 +[16] Hezhen Hu, Wengang Zhou, Junfu Pu, and Houqiang Li. Global-local enhancement network for nmf-aware sign language recognition. ACM transactions on multimedia computing, communications, and applications (TOMM), 17(3): 1-19, 2021. 2, 7 +[17] Hezhen Hu, Weichao Zhao, Wengang Zhou, and Houqiang Li. Signbert+: Hand-model-aware self-supervised pretraining for sign language understanding. IEEE TPAMI, 45 (9):11221-11239, 2023. 6 +[18] Lianyu Hu, Liqing Gao, Zekang Liu, and Wei Feng. Dynamic spatial-temporal aggregation for skeleton-aware sign language recognition. In LREC/COLING, 2024. 6, 7 +[19] Jie Huang, Wengang Zhou, Houqiang Li, and Weiping Li. Attention-based 3d-cnns for large-vocabulary sign language recognition. IEEE TCSVT, 29(9):2822-2832, 2018. 2 +[20] Jie Huang, Wengang Zhou, Qilin Zhang, Houqiang Li, and Weiping Li. Video-based sign language recognition without temporal segmentation. In AAAI, 2018. 2 +[21] Songyao Jiang, Bin Sun, Lichen Wang, Yue Bai, Kunpeng Li, and Yun Fu. Skeleton aware multi-modal sign language recognition. In CVPR, pages 3413-3423, 2021. 1, 3, 4, 5, 8 +[22] Songyao Jiang, Bin Sun, Lichen Wang, Yue Bai, Kunpeng Li, and Yun Fu. Sign language recognition via skeleton-aware multi-model ensemble. arXiv preprint arXiv:2110.06161, 2021. 6, 7 +[23] Tao Jiang, Necati Cihan Camgoz, and Richard Bowden. Skeleton: Skeletal transformers for robust body-posed estimation. In CVPR, pages 3394–3402, 2021. 1, 3 +[24] Sheng Jin, Lumin Xu, Jin Xu, Can Wang, Wentao Liu, Chen Qian, Wanli Ouyang, and Ping Luo. Whole-body human pose estimation in the wild. In ECCV, pages 196-214. Springer, 2020. 3, 4 +[25] Trevor Johnston and Adam Schembri. Australian Sign Language (Auslan): An introduction to sign language linguistics. Cambridge University Press, 2007. 1 +[26] Hamid Reza Vaezi Joze and Oscar Koller. Ms-asl: A large-scale data set and benchmark for understanding american sign language. arXiv preprint arXiv:1812.01053, 2018. 2, 7 +[27] Alexander Kapitanov, Kvanchiani Karina, Alexander Nagaev, and Petrova Elizaveta. Slovo: Russian sign language dataset. In International Conference on Computer Vision Systems, pages 63-73. Springer, 2023. 2, 7, 8 + +[28] Oscar Koller. Quantitative survey of the state of the art in sign language recognition. arXiv preprint arXiv:2008.09918, 2020. 2 +[29] Oscar Koller, Sepehr Zargaran, Hermann Ney, and Richard Bowden. Deep sign: Enabling robust statistical continuous sign language recognition via hybrid cnn-hmms. *IJCV*, 126: 1311–1325, 2018. 2 +[30] Taeryung Lee, Yeonguk Oh, and Kyoung Mu Lee. Human part-wise 3d motion context learning for sign language recognition. In ICCV, pages 20740-20750, 2023. 1, 6 +[31] Chao Li, Qiaoyong Zhong, Di Xie, and Shiliang Pu. Cooccurrence feature learning from skeleton data for action recognition and detection with hierarchical aggregation. arXiv preprint arXiv:1804.06055, 2018. 1, 3 +[32] Dongxu Li, Cristian Rodriguez, Xin Yu, and Hongdong Li. Word-level deep sign language recognition from video: A new large-scale dataset and methods comparison. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 1459-1469, 2020. 2, 7, 8 +[33] Dongxu Li, Xin Yu, Chenchen Xu, Lars Petersson, and Hongdong Li. Transferring cross-domain knowledge for video sign language recognition. In CVPR, pages 6205-6214, 2020. 6 +[34] Maosen Li, Siheng Chen, Xu Chen, Ya Zhang, Yanfeng Wang, and Qi Tian. Actional-structural graph convolutional networks for skeleton-based action recognition. In CVPR, pages 3595–3603, 2019. 3 +[35] Yanghao Li, Chao-Yuan Wu, Haoqi Fan, Karttikeya Mangalam, Bo Xiong, Jitendra Malik, and Christoph Feichtenhofer. Mviv2: Improved multiscale vision transformers for classification and detection. In CVPR, pages 4804-4814, 2022. 7, 8 +[36] Jiajia Liao, Yujun Liu, Yingchao Piao, Jinhe Su, Guorong Cai, and Yundong Wu. Gle-net: A global and local ensemble network for aerial object detection. International Journal of Computational Intelligence Systems, 15(1):2, 2022. 7 +[37] Jinfu Liu, Xinshun Wang, Can Wang, Yuan Gao, and Mengyuan Liu. Temporal decoupling graph convolutional network for skeleton-based gesture recognition. IEEE TMM, 26:811-823, 2023. 3 +[38] Ze Liu, Jia Ning, Yue Cao, Yixuan Wei, Zheng Zhang, Stephen Lin, and Han Hu. Video swim transformer. In CVPR, pages 3202-3211, 2022. 3, 7, 8 +[39] Anna Mindess. Reading between the signs: Intercultural communication for sign language interpreters. Nicholas Brealey, 2014. 1 +[40] Liliane Momeni, Hannah Bull, KR Prajwal, Samuel Albanie, Gül Varol, and Andrew Zisserman. Automatic dense annotation of large-vocabulary sign language videos. In ECCV, pages 671-690. Springer, 2022. 2 +[41] Evonne Ng, Shiry Ginosar, Trevor Darrell, and Hanbyul Joo. Body2hands: Learning to infer 3d hands from conversational gesture body dynamics. In CVPR, pages 11865-11874, 2021. 1, 3 +[42] Ladislav Rampášek, Michael Galkin, Vijay Prakash Dwivedi, Anh Tuan Luu, Guy Wolf, and Dominique Beaini. Recipe for a general, powerful, scalable graph transformer. NeurIPS, 35:14501-14515, 2022. 3 + +[43] Razieh Rastgoo, Kourosh Kiani, and Sergio Escalera. Sign language recognition: A deep survey. Expert Systems with Applications, 164:113794, 2021. 2 +[44] Wendy Sandler. Sign language and linguistic universals. Cambridge University, 2006. 1 +[45] Noha Sarhan and Simone Frintrop. Unraveling a decade: a comprehensive survey on isolated sign language recognition. In ICCV, pages 3210-3219, 2023. 2 +[46] Lei Shi, Yifan Zhang, Jian Cheng, and Hanqing Lu. Two-stream adaptive graph convolutional networks for skeleton-based action recognition. In CVPR, pages 12026-12035, 2019. 3 +[47] Lei Shi, Yifan Zhang, Jian Cheng, and Hanqing Lu. Skeleton-based action recognition with multi-stream adaptive graph convolutional networks. IEEE TIP, 29:9532-9545, 2020. 2 +[48] Chenyang Si, Wentao Chen, Wei Wang, Liang Wang, and Tieniu Tan. An attention enhanced graph convolutional LSTM network for skeleton-based action recognition. In CVPR, pages 1227-1236, 2019. 3 +[49] Neil Song and Yu Xiang. Slgtformer: An attention-based approach to sign language recognition. arXiv preprint arXiv:2212.10746, 2022. 3, 4 +[50] Yi-Fan Song, Zhang Zhang, Caifeng Shan, and Liang Wang. Stronger, faster and more explainable: A graph convolutional baseline for skeleton-based action recognition. In proceedings of the 28th ACM international conference on multimedia, pages 1625-1633, 2020. 2 +[51] Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. Deep high-resolution representation learning for human pose estimation. In CVPR, pages 5693-5703, 2019. 3, 4 +[52] Clayton Valli. Linguistics of American Sign Language: An Introduction. Gallaudet University Press, 2000. 1 +[53] Gül Varol, Liliane Momeni, Samuel Albanie, Triantafyllos Afouras, and Andrew Zisserman. Scaling up sign spotting through sign language dictionaries. *IJCV*, 130(6):1416–1439, 2022. 2 +[54] Hongsong Wang and Liang Wang. Modeling temporal dynamics and spatial configurations of actions using two-stream recurrent neural networks. In CVPR, pages 499-508, 2017. 2 +[55] Hanjie Wang, Xiujuan Chai, and Xin Chen. A novel sign language recognition framework using hierarchical grassmann covariance matrix. IEEE TMM, 21(11):2806-2814, 2019. 2 +[56] Chengcheng Wei, Jian Zhao, Wengang Zhou, and Houqiang Li. Semantic boundary detection with reinforcement learning for continuous sign language recognition. IEEE TCSVT, 31(3):1138-1149, 2020. +[57] Pan Xie, Mengyi Zhao, and Xiaohui Hu. Pisltrc: Position-informed sign language transformer with content-aware convolution. IEEE TMM, 24:3908-3919, 2021. 2 +[58] Sijie Yan, Yuanjun Xiong, and Dahua Lin. Spatial temporal graph convolutional networks for skeleton-based action recognition. In AAAI, 2018. 1, 6, 7 +[59] Aoxiong Yin, Zhou Zhao, Weike Jin, Meng Zhang, Xingshan Zeng, and Xiaofei He. Mlslt: Towards multilingual sign language translation. In CVPR, pages 5109-5119, 2022. 1 + +[60] Weichao Zhao, Hezhen Hu, Wengang Zhou, Jiaxin Shi, and Houqiang Li. Best: Bert pre-training for sign language recognition with coupling tokenization. In AAAI, pages 3597-3605, 2023. 6, 7 +[61] Weichao Zhao, Hezhen Hu, Wengang Zhou, Yunyao Mao, Min Wang, and Houjiang Li. Masa: Motion-aware masked autoencoder with semantic alignment for sign language recognition. IEEE TCSVT, 2024. 6, 7 +[62] Weichao Zhao, Wengang Zhou, Hezhen Hu, Min Wang, and Houqiang Li. Self-supervised representation learning with spatial-temporal consistency for sign language recognition. arXiv preprint arXiv:2406.10501, 2024. 6, 7 +[63] Ronglai Zuo, Fangyun Wei, and Brian Mak. Natural language-assisted sign language recognition. In CVPR, pages 14890-14900, 2023. 6, 7 \ No newline at end of file diff --git a/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/images.zip b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/images.zip new file mode 100644 index 0000000000000000000000000000000000000000..0427925cce90e519343dfd9a91e75eb0b06a8c5a --- /dev/null +++ b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/images.zip @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9683809fddb618d4fe8a5df52fe9ed388309dc1ec835c8a11206d3894471c8e6 +size 680534 diff --git a/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/layout.json b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/layout.json new file mode 100644 index 0000000000000000000000000000000000000000..3e951aa0dfad42b7eb8ee71eb68d869c26f640fd --- /dev/null +++ b/2025/VSNet_ Focusing on the Linguistic Characteristics of Sign Language/layout.json @@ -0,0 +1,8751 @@ +{ + "pdf_info": [ + { + "para_blocks": [ + { + "bbox": [ + 94, + 103, + 516, + 121 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 94, + 103, + 516, + 121 + ], + "spans": [ + { + "bbox": [ + 94, + 103, + 516, + 121 + ], + "type": "text", + "content": "VSNet: Focusing on the Linguistic Characteristics of Sign Language" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "spans": [ + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "text", + "content": "Yuhao Li" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "inline_equation", + "content": "^{1}" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "text", + "content": ", Xinyue Chen" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "inline_equation", + "content": "^{1}" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "text", + "content": ", Hongkai Li" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "inline_equation", + "content": "^{1}" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "text", + "content": ", Xiaorong Pu" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "inline_equation", + "content": "^{1,3}" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "text", + "content": ", Peng Jin" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "inline_equation", + "content": "^{2}" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "text", + "content": ", Yazhou Ren" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "inline_equation", + "content": "^{1,3,*}" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "inline_equation", + "content": "^{1}" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "text", + "content": "School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China; " + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "inline_equation", + "content": "^{2}" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "text", + "content": "Sichuan Provincial Key Lab of Philosophy and Social Science for Language Intelligence in Special Education, Leshan Normal University, Leshan, China; " + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "inline_equation", + "content": "^{3}" + }, + { + "bbox": [ + 96, + 142, + 529, + 215 + ], + "type": "text", + "content": "Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China, Shenzhen, China" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 151, + 241, + 200, + 253 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 151, + 241, + 200, + 253 + ], + "spans": [ + { + "bbox": [ + 151, + 241, + 200, + 253 + ], + "type": "text", + "content": "Abstract" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 266, + 296, + 542 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 266, + 296, + 542 + ], + "spans": [ + { + "bbox": [ + 55, + 266, + 296, + 542 + ], + "type": "text", + "content": "Sign language is a visual language expressed through complex movements of the upper body. The human skeleton plays a critical role in sign language recognition due to its good separation from the video background. However, mainstream skeleton-based sign language recognition models often overly focus on the natural connections between joints, treating sign language as ordinary human movements, which neglects its linguistic characteristics. We believe that just as letters form words, each sign language gloss can also be decomposed into smaller visual symbols. To fully harness the potential of skeleton data, this paper proposes a novel joint fusion strategy and a visual symbol attention model. Specifically, we first input the complete set of skeletal joints, and after dynamically exchanging joint information, we discard the parts with the weakest connections to other joints, resulting in a fused, simplified skeleton. Then, we group the joints most likely to express the same visual symbol and discuss the joint movements within each group separately. To validate the superiority of our method, we conduct extensive experiments on multiple public benchmark datasets. The results show that, without complex pretraining, we still achieve new state-of-the-art performance. The code is available at https://github.com/atinyboy/VSNet." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 56, + 578, + 135, + 590 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 578, + 135, + 590 + ], + "spans": [ + { + "bbox": [ + 56, + 578, + 135, + 590 + ], + "type": "text", + "content": "1. Introduction" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 55, + 597, + 295, + 694 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 597, + 295, + 694 + ], + "spans": [ + { + "bbox": [ + 55, + 597, + 295, + 694 + ], + "type": "text", + "content": "Sign language is a visual language expressed through body movements and serves as the primary communication tool for the deaf community [13]. Although sign language and spoken language have distinctly different grammatical structures [1, 7], they share common linguistic features [39, 44, 59]. Accurate sign language translation can facilitate communication between deaf and hearing people, fostering a more inclusive and accessible society. Among" + } + ] + } + ], + "index": 7 + }, + { + "type": "image", + "bbox": [ + 320, + 240, + 376, + 297 + ], + "blocks": [ + { + "bbox": [ + 320, + 240, + 376, + 297 + ], + "lines": [ + { + "bbox": [ + 320, + 240, + 376, + 297 + ], + "spans": [ + { + "bbox": [ + 320, + 240, + 376, + 297 + ], + "type": "image", + "image_path": "0ef7164ed9cc0e33de67ef10e8993b7a4a21ae4ff844e5a166b44fb3d111af05.jpg" + } + ] + } + ], + "index": 8, + "angle": 0, + "type": "image_body" + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 378, + 240, + 434, + 297 + ], + "blocks": [ + { + "bbox": [ + 378, + 240, + 434, + 297 + ], + "lines": [ + { + "bbox": [ + 378, + 240, + 434, + 297 + ], + "spans": [ + { + "bbox": [ + 378, + 240, + 434, + 297 + ], + "type": "image", + "image_path": "44fe15157fcbe074d6c1d7cd80d06ce1f7f18531872187c473e9164ff0c00f43.jpg" + } + ] + } + ], + "index": 9, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 416, + 299, + 456, + 307 + ], + "lines": [ + { + "bbox": [ + 416, + 299, + 456, + 307 + ], + "spans": [ + { + "bbox": [ + 416, + 299, + 456, + 307 + ], + "type": "text", + "content": "gloss: 'accent'" + } + ] + } + ], + "index": 12, + "angle": 0, + "type": "image_caption" + } + ], + "index": 9 + }, + { + "type": "image", + "bbox": [ + 435, + 240, + 492, + 297 + ], + "blocks": [ + { + "bbox": [ + 435, + 240, + 492, + 297 + ], + "lines": [ + { + "bbox": [ + 435, + 240, + 492, + 297 + ], + "spans": [ + { + "bbox": [ + 435, + 240, + 492, + 297 + ], + "type": "image", + "image_path": "f9bf732e363bf30348c2d295f318f5d05c852e71aa7b797239453d99c7a09b96.jpg" + } + ] + } + ], + "index": 10, + "angle": 0, + "type": "image_body" + } + ], + "index": 10 + }, + { + "type": "image", + "bbox": [ + 493, + 240, + 549, + 297 + ], + "blocks": [ + { + "bbox": [ + 493, + 240, + 549, + 297 + ], + "lines": [ + { + "bbox": [ + 493, + 240, + 549, + 297 + ], + "spans": [ + { + "bbox": [ + 493, + 240, + 549, + 297 + ], + "type": "image", + "image_path": "ca5a47bc0d09f7265a256a59d8607956570ca3711d2ec5e38228d7f2796734ad.jpg" + } + ] + } + ], + "index": 11, + "angle": 0, + "type": "image_body" + } + ], + "index": 11 + }, + { + "type": "image", + "bbox": [ + 320, + 314, + 376, + 370 + ], + "blocks": [ + { + "bbox": [ + 320, + 314, + 376, + 370 + ], + "lines": [ + { + "bbox": [ + 320, + 314, + 376, + 370 + ], + "spans": [ + { + "bbox": [ + 320, + 314, + 376, + 370 + ], + "type": "image", + "image_path": "fdeac63056da7ab56e05b49bb7c6ccfa708b60992d811714809658431cc4ca36.jpg" + } + ] + } + ], + "index": 13, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 313, + 388, + 555, + 444 + ], + "lines": [ + { + "bbox": [ + 313, + 388, + 555, + 444 + ], + "spans": [ + { + "bbox": [ + 313, + 388, + 555, + 444 + ], + "type": "text", + "content": "Figure 1. In the sign for \"accent\", the index and middle fingers (or only index finger) are extended toward the neck while other fingers are curled, forming the \"pointing\" VS. In the sign for \"algebra\", movements can be decomposed into multiple VS such as \"crossed arms,\" \"fist,\" and \"thumb-up.\"" + } + ] + } + ], + "index": 18, + "angle": 0, + "type": "image_caption" + } + ], + "index": 13 + }, + { + "type": "image", + "bbox": [ + 378, + 314, + 433, + 370 + ], + "blocks": [ + { + "bbox": [ + 378, + 314, + 433, + 370 + ], + "lines": [ + { + "bbox": [ + 378, + 314, + 433, + 370 + ], + "spans": [ + { + "bbox": [ + 378, + 314, + 433, + 370 + ], + "type": "image", + "image_path": "18fa9e3dbae1a5067929bcacce7062c5be4971a7cddd1ad711e479f11e05de4b.jpg" + } + ] + } + ], + "index": 14, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 416, + 372, + 455, + 380 + ], + "lines": [ + { + "bbox": [ + 416, + 372, + 455, + 380 + ], + "spans": [ + { + "bbox": [ + 416, + 372, + 455, + 380 + ], + "type": "text", + "content": "gloss:'algebra'" + } + ] + } + ], + "index": 17, + "angle": 0, + "type": "image_caption" + } + ], + "index": 14 + }, + { + "type": "image", + "bbox": [ + 435, + 314, + 492, + 370 + ], + "blocks": [ + { + "bbox": [ + 435, + 314, + 492, + 370 + ], + "lines": [ + { + "bbox": [ + 435, + 314, + 492, + 370 + ], + "spans": [ + { + "bbox": [ + 435, + 314, + 492, + 370 + ], + "type": "image", + "image_path": "ad3e18e00f9ec6d67c41252c0742477cc98782276705a52e360e368fa82d5f19.jpg" + } + ] + } + ], + "index": 15, + "angle": 0, + "type": "image_body" + } + ], + "index": 15 + }, + { + "type": "image", + "bbox": [ + 493, + 314, + 548, + 370 + ], + "blocks": [ + { + "bbox": [ + 493, + 314, + 548, + 370 + ], + "lines": [ + { + "bbox": [ + 493, + 314, + 548, + 370 + ], + "spans": [ + { + "bbox": [ + 493, + 314, + 548, + 370 + ], + "type": "image", + "image_path": "10785b8de7ae97caba6d78a7d9b114606070865a8f7c70ff8eea76801732b85a.jpg" + } + ] + } + ], + "index": 16, + "angle": 0, + "type": "image_body" + } + ], + "index": 16 + }, + { + "bbox": [ + 313, + 471, + 555, + 532 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 471, + 555, + 532 + ], + "spans": [ + { + "bbox": [ + 313, + 471, + 555, + 532 + ], + "type": "text", + "content": "the key subtasks in sign language translation is isolated sign language recognition (ISLR), which focuses on translating individual glosses. ISLR has garnered significant attention as it aids people in accurately understanding numerous gestures and quickly learning sign language." + } + ] + } + ], + "index": 19 + }, + { + "bbox": [ + 313, + 534, + 556, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 534, + 556, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 534, + 556, + 715 + ], + "type": "text", + "content": "In recent years, human skeletons have been increasingly applied in the ISLR field due to their strong operability and clear separation from video backgrounds. Although most skeleton-based recognition methods have achieved promising results, they typically treat sign language as typical human motion [12, 21, 23, 30, 31, 41, 58], overlooking the intrinsic relationships among joints and the unique linguistic characteristics [25, 52] of sign language itself. We believe that: (1) The importance of joints in sign language is not equivalent; joints with stronger connections to others are more significant than isolated joints, and connections among joints with similar movement tendencies hold more meaning than those among dispersed joints. (2) Similar to how letters form words, each sign language gloss can be decomposed into smaller visual symbols (VS). While VS" + } + ] + } + ], + "index": 20 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "spans": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "type": "text", + "content": "CVF" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 145, + 0, + 496, + 37 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 145, + 0, + 496, + 37 + ], + "spans": [ + { + "bbox": [ + 145, + 0, + 496, + 37 + ], + "type": "text", + "content": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 66, + 702, + 234, + 713 + ], + "type": "page_footnote", + "angle": 0, + "lines": [ + { + "bbox": [ + 66, + 702, + 234, + 713 + ], + "spans": [ + { + "bbox": [ + 66, + 702, + 234, + 713 + ], + "type": "text", + "content": "*Corresponding author (yazhou.ren@uestc.edu.cn)." + } + ] + } + ], + "index": 21 + }, + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "text", + "content": "24320" + } + ] + } + ], + "index": 22 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 0 + }, + { + "para_blocks": [ + { + "bbox": [ + 55, + 72, + 295, + 120 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 72, + 295, + 120 + ], + "spans": [ + { + "bbox": [ + 55, + 72, + 295, + 120 + ], + "type": "text", + "content": "lack explicit meaning on their own, from a skeletal perspective, they capture distinct movement patterns of joints, effectively decomposing sign language actions (see Fig. 1). Therefore, our work focuses on these two core aspects:" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 122, + 295, + 361 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 122, + 295, + 361 + ], + "spans": [ + { + "bbox": [ + 56, + 122, + 295, + 361 + ], + "type": "text", + "content": "Core 1: Focus on identifying essential joints. Including too many joints can impede training, while too few can lead to information loss. On the one hand, many prior models have used manually trimmed, simplified skeletons, but manually selecting the optimal joints is not trivial. On the other hand, enabling a model to automatically choose key joints can be computationally intensive. Fortunately, the dynamically learned skeletons from GCNs approximate real structures more closely, enabling us to reduce the dependency on an initial skeleton. By computing the weight of each joint, we can identify and retain the most important joints while discarding less critical local joints. Specifically, for the complete hand skeleton, we discard a small number of joints based on computed weights after each GCN layer, relearning new skeleton connections in the subsequent GCN layer. We control the number of joints discarded within each region to ensure that the final GCN output retains a recognizable structure. Notably, even joints with the lowest weights undergo at least one convolution, so their information is not entirely discarded." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 363, + 295, + 567 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 363, + 295, + 567 + ], + "spans": [ + { + "bbox": [ + 56, + 363, + 295, + 567 + ], + "type": "text", + "content": "Core 2: Focus on emphasizing visual symbols (VS) in sign language. If each joint is considered the smallest element of a sign language action, then the combination of joints across all frames becomes too vast to compute self-attention. In practice, sign language is not random motion; it is designed for human recognition. When a signer interprets sign language, they do not scrutinize the position of every joint but instead form a general impression of each hand. Thus, in this work, we concentrate on recognizing VS. Unfortunately, isolating each VS from movements is as challenging as the entire SLR task itself. Therefore, we adopt a simplified approach, segmenting the skeleton into a generalized VS in a fixed manner, as shown in Fig. 2. We then design two self-attention modules to capture spatial and temporal relationships between these VS. Empirically, we find that this approach significantly enhances the model's discriminative power." + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 67, + 568, + 275, + 579 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 67, + 568, + 275, + 579 + ], + "spans": [ + { + "bbox": [ + 67, + 568, + 275, + 579 + ], + "type": "text", + "content": "Our main contributions are summarized as follows:" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 582, + 295, + 713 + ], + "type": "list", + "angle": 0, + "index": 7, + "blocks": [ + { + "bbox": [ + 55, + 582, + 295, + 641 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 582, + 295, + 641 + ], + "spans": [ + { + "bbox": [ + 55, + 582, + 295, + 641 + ], + "type": "text", + "content": "- We propose a novel module with an automated skeleton simplification function for sign language recognition. This module is designed to dynamically learn global skeleton connections while progressively discarding isolated local joints." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 642, + 295, + 700 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 642, + 295, + 700 + ], + "spans": [ + { + "bbox": [ + 55, + 642, + 295, + 700 + ], + "type": "text", + "content": "- Considering the linguistic characteristics of sign language, we introduce a new self-attention model called VSformer, which captures temporal and spatial patterns in sign language motions and extracts knowledge from the human skeleton based on the concept of VS." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 701, + 295, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 701, + 295, + 713 + ], + "spans": [ + { + "bbox": [ + 55, + 701, + 295, + 713 + ], + "type": "text", + "content": "- We conduct extensive experiments on ISLR to validate" + } + ] + } + ], + "index": 6 + } + ], + "sub_type": "text" + }, + { + "type": "image", + "bbox": [ + 320, + 72, + 547, + 203 + ], + "blocks": [ + { + "bbox": [ + 320, + 72, + 547, + 203 + ], + "lines": [ + { + "bbox": [ + 320, + 72, + 547, + 203 + ], + "spans": [ + { + "bbox": [ + 320, + 72, + 547, + 203 + ], + "type": "image", + "image_path": "643f5b9a8f8824d4be18635ddea1b3f7e73bb10c6cc91a7c3c93649d5a9787aa.jpg" + } + ] + } + ], + "index": 8, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 317, + 213, + 550, + 224 + ], + "lines": [ + { + "bbox": [ + 317, + 213, + 550, + 224 + ], + "spans": [ + { + "bbox": [ + 317, + 213, + 550, + 224 + ], + "type": "text", + "content": "Figure 2. Decomposing generalized visual symbols from videos" + } + ] + } + ], + "index": 9, + "angle": 0, + "type": "image_caption" + } + ], + "index": 8 + }, + { + "bbox": [ + 321, + 246, + 553, + 294 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 321, + 246, + 553, + 294 + ], + "spans": [ + { + "bbox": [ + 321, + 246, + 553, + 294 + ], + "type": "text", + "content": "the effectiveness of our proposed approach. Compared to previous methods, our approach achieves SoTA performance on four public benchmarks, namely WLASL [32], MSASL [26], NMFs-CSR [16], and Slovo [27]." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 314, + 304, + 406, + 316 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 304, + 406, + 316 + ], + "spans": [ + { + "bbox": [ + 314, + 304, + 406, + 316 + ], + "type": "text", + "content": "2. Related Works" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 314, + 324, + 465, + 337 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 324, + 465, + 337 + ], + "spans": [ + { + "bbox": [ + 314, + 324, + 465, + 337 + ], + "type": "text", + "content": "2.1. Sign Language Recognition" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 313, + 342, + 555, + 533 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 342, + 555, + 533 + ], + "spans": [ + { + "bbox": [ + 313, + 342, + 555, + 533 + ], + "type": "text", + "content": "Sign language is the primary means of communication within the deaf community. With significant advances in machine learning and computer vision over the past decade, exploring sign language recognition (SLR) has become increasingly important [28, 43, 45], as it can automatically interpret sign language and help deaf and mute individuals communicate more easily in daily life. Unlike directly translating sign language sentences, isolated sign language recognition (ISLR) is a subtask in the field of sign language translation [45], aimed at translating each individual gloss [40, 53]. ISLR is not only useful for guiding beginners in learning sign language and assisting in the creation of new sign language gestures, but it also lays a solid foundation for SLR [1, 4, 53]. Research in this area can be divided into two categories based on the input modality: RGB-based methods and skeleton-based methods." + } + ] + } + ], + "index": 13 + }, + { + "bbox": [ + 313, + 534, + 555, + 629 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 534, + 555, + 629 + ], + "spans": [ + { + "bbox": [ + 313, + 534, + 555, + 629 + ], + "type": "text", + "content": "RGB-based Recognition. RGB-based methods, similar to other video recognition tasks, typically use CNNs to extract visual features from RGB videos [5, 15, 19, 20, 29, 55-57], and then aggregate temporal and spatial information to predict sign language. However, RGB-based models face two significant challenges: first, they are heavily affected by changes in the video background, and second, video data occupies a large amount of memory resources." + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 313, + 630, + 555, + 714 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 630, + 555, + 714 + ], + "spans": [ + { + "bbox": [ + 313, + 630, + 555, + 714 + ], + "type": "text", + "content": "Skeleton-based Recognition. Compared to RGB images, human skeletons [8, 47, 50, 54] describe the motion information conveyed in sign language videos using simplified points and edges, reducing interference from complex backgrounds and the appearance of the signer. Moreover, operations on three-dimensional spatial points with actual physical meanings offer stronger interpretability and" + } + ] + } + ], + "index": 15 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "24321" + } + ] + } + ], + "index": 16 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 1 + }, + { + "para_blocks": [ + { + "type": "image", + "bbox": [ + 61, + 70, + 545, + 274 + ], + "blocks": [ + { + "bbox": [ + 61, + 70, + 545, + 274 + ], + "lines": [ + { + "bbox": [ + 61, + 70, + 545, + 274 + ], + "spans": [ + { + "bbox": [ + 61, + 70, + 545, + 274 + ], + "type": "image", + "image_path": "a0fe65ed0df5a97890f903093b93c51daf44c0b86f7de44230e2f011d4ddd927.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 59, + 281, + 550, + 315 + ], + "lines": [ + { + "bbox": [ + 59, + 281, + 550, + 315 + ], + "spans": [ + { + "bbox": [ + 59, + 281, + 550, + 315 + ], + "type": "text", + "content": "Figure 3. Overview of our VSNet. Here, Down sample consists of a convolution layer with a kernel size of 7 and a stride of 2 along the temporal dimension, followed by a batch norm layer, Average Pooling represent transforming the data into a single-dimensional feature by taking the mean across both temporal and spatial dimensions. For details on the Skeleton-" + }, + { + "bbox": [ + 59, + 281, + 550, + 315 + ], + "type": "inline_equation", + "content": "(a,b,c)" + }, + { + "bbox": [ + 59, + 281, + 550, + 315 + ], + "type": "text", + "content": ", please refer to Fig. 4." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "image_caption" + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 335, + 296, + 396 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 335, + 296, + 396 + ], + "spans": [ + { + "bbox": [ + 55, + 335, + 296, + 396 + ], + "type": "text", + "content": "lower memory consumption. In earlier works, identifying the correct human keypoints in a video was a major challenge. However, with the development of deep learning, human pose estimation models have become increasingly reliable [10, 24, 51]." + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 405, + 157, + 417 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 405, + 157, + 417 + ], + "spans": [ + { + "bbox": [ + 55, + 405, + 157, + 417 + ], + "type": "text", + "content": "2.2. Human Skeleton" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 54, + 424, + 295, + 628 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 424, + 295, + 628 + ], + "spans": [ + { + "bbox": [ + 54, + 424, + 295, + 628 + ], + "type": "text", + "content": "Due to the natural physical connections between skeleton joints, Graph Convolutional Network (GCN) has been widely applied to model spatio-temporal relationships [12, 23, 31, 34, 41, 46]. GCN require predefined connectivity of the key points to construct the graph network. However, the physically natural skeleton connections may not always be the most suitable edges for recognition. Relying solely on the skeleton itself to construct the graph can limit the information flow between nodes. To address this issue, SL-GCN [21] employs a dynamic graph with decoupling between different frames, while TD-GCN [37] adds supernodes to control the connections between nodes. A common drawback of these approaches is that they require complex networks to model the key points. Furthermore, while dynamic graphs alleviate the limitations of the original skeleton, blindly allowing all key points to connect introduces redundant information, leading to suboptimal models." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 628, + 296, + 653 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 628, + 296, + 653 + ], + "spans": [ + { + "bbox": [ + 55, + 628, + 296, + 653 + ], + "type": "text", + "content": "Therefore, we argue that relying solely on GCN does not fully exploit the true potential of skeleton data." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 654, + 296, + 714 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 654, + 296, + 714 + ], + "spans": [ + { + "bbox": [ + 55, + 654, + 296, + 714 + ], + "type": "text", + "content": "In recent years, attention-based methods [3, 6, 9, 38, 42, 48] have also made progress in human skeleton recognition. Unlike GCN, attention mechanisms do not require the researcher to provide information on joint connections, instead equally attending to all nodes. However, captur" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 313, + 335, + 555, + 478 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 335, + 555, + 478 + ], + "spans": [ + { + "bbox": [ + 313, + 335, + 555, + 478 + ], + "type": "text", + "content": "ing the correlations between all joints across all frames requires substantial memory resources. To address this, SLGTformer [49] decomposes the spatio-temporal pose sequence into spatial graphs and temporal windows, while SkateFormer [11] partitions joints and frames and performs self-attention on each partition. Based on these theories, we design a model where GCNs and self-attention networks complement each other. GCNs are responsible for exploring the spatial correlations between human joints, while self-attention networks focus on capturing the motion of the skeleton across multiple time sequences and spatial combinations." + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 314, + 493, + 375, + 504 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 493, + 375, + 504 + ], + "spans": [ + { + "bbox": [ + 314, + 493, + 375, + 504 + ], + "type": "text", + "content": "3. Methods" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 313, + 513, + 432, + 527 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 513, + 432, + 527 + ], + "spans": [ + { + "bbox": [ + 313, + 513, + 432, + 527 + ], + "type": "text", + "content": "3.1. Data Pre-Processing" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 533, + 555, + 675 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 533, + 555, + 675 + ], + "spans": [ + { + "bbox": [ + 313, + 533, + 555, + 675 + ], + "type": "text", + "content": "Key Point Extraction. Sign language is a visual language primarily conveyed through hand and body movements, with facial expressions also playing a crucial role. However, the human skeleton mainly describes bones and joints, excluding muscles, making it unable to accurately capture facial expressions in sign language videos. Our investigation revealed that a significant number of signers, especially beginners, do not exhibit extensive facial expressions during sign language communication. Therefore, our main experiments exclude facial expressions, while the appendix provides additional details on how the concept of VS can be extended to the face." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "text", + "content": "In previous GCN models, finger joints were often simplified to avoid making the graph too complex, leading to a loss of information. In this study, we selected a complete" + } + ] + } + ], + "index": 11 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "text", + "content": "24322" + } + ] + } + ], + "index": 12 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 2 + }, + { + "para_blocks": [ + { + "type": "image", + "bbox": [ + 58, + 72, + 358, + 171 + ], + "blocks": [ + { + "bbox": [ + 58, + 72, + 358, + 171 + ], + "lines": [ + { + "bbox": [ + 58, + 72, + 358, + 171 + ], + "spans": [ + { + "bbox": [ + 58, + 72, + 358, + 171 + ], + "type": "image", + "image_path": "3b302e9367d27a69b446d6fd883fe21af9ef8d9b70365c49823d0a11074b393e.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 55, + 182, + 364, + 228 + ], + "lines": [ + { + "bbox": [ + 55, + 182, + 364, + 228 + ], + "spans": [ + { + "bbox": [ + 55, + 182, + 364, + 228 + ], + "type": "text", + "content": "Figure 4. Illustration of the skeleton and grouping type-1. " + }, + { + "bbox": [ + 55, + 182, + 364, + 228 + ], + "type": "inline_equation", + "content": "V_{0}" + }, + { + "bbox": [ + 55, + 182, + 364, + 228 + ], + "type": "text", + "content": " represents a localization node, the dashed box indicates nodes that are grouped together and hollow dots indicate dropped nodes. Please note that (a) is the input real skeleton diagram, while (b) and (c) are simplified representations drawn after dropping for ease of analysis." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "image_caption" + } + ], + "index": 0 + }, + { + "type": "image", + "bbox": [ + 380, + 72, + 550, + 186 + ], + "blocks": [ + { + "bbox": [ + 380, + 72, + 550, + 186 + ], + "lines": [ + { + "bbox": [ + 380, + 72, + 550, + 186 + ], + "spans": [ + { + "bbox": [ + 380, + 72, + 550, + 186 + ], + "type": "image", + "image_path": "b715aea289b6efb25c9f8e0f180402f2a4f181c1c5aefe14eb1470b89cdc6b34.jpg" + } + ] + } + ], + "index": 2, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 376, + 201, + 555, + 224 + ], + "lines": [ + { + "bbox": [ + 376, + 201, + 555, + 224 + ], + "spans": [ + { + "bbox": [ + 376, + 201, + 555, + 224 + ], + "type": "text", + "content": "Figure 5. Skeleton groupings type-2 and type-3. The other hand is grouped in a mirrored fashion." + } + ] + } + ], + "index": 3, + "angle": 0, + "type": "image_caption" + } + ], + "index": 2 + }, + { + "bbox": [ + 54, + 247, + 294, + 283 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 247, + 294, + 283 + ], + "spans": [ + { + "bbox": [ + 54, + 247, + 294, + 283 + ], + "type": "text", + "content": "set of 44 key points from both hands out of the 133 human body joints extracted using HRNet [24, 51], as shown in Fig. 4(a)." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 54, + 283, + 295, + 415 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 283, + 295, + 415 + ], + "spans": [ + { + "bbox": [ + 54, + 283, + 295, + 415 + ], + "type": "text", + "content": "Data Enhancing. Given a video " + }, + { + "bbox": [ + 54, + 283, + 295, + 415 + ], + "type": "inline_equation", + "content": "X \\in \\mathbb{R}^{T \\times V \\times C}" + }, + { + "bbox": [ + 54, + 283, + 295, + 415 + ], + "type": "text", + "content": ", where T, V, and C represent the number of frames, joints, and coordinates respectively. We replace abnormal points—such as those out of range (less than 0, more than 512) or clearly distorted—caused by model extraction errors, with the average value from adjacent frames. Additionally, to standardize body positions and movement amplitudes across different signers in the sign language videos, we first use the hollow node " + }, + { + "bbox": [ + 54, + 283, + 295, + 415 + ], + "type": "inline_equation", + "content": "V_{0}" + }, + { + "bbox": [ + 54, + 283, + 295, + 415 + ], + "type": "text", + "content": " in Fig. 4 as the reference node for the entire video. Then, all key points in each frame are shifted based on this reference coordinate, as follows:" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 80, + 427, + 295, + 441 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 80, + 427, + 295, + 441 + ], + "spans": [ + { + "bbox": [ + 80, + 427, + 295, + 441 + ], + "type": "interline_equation", + "content": "X (t, v, c) = X (t, v, c) - X (t, v _ {0}, c) + [ x, y, z ] \\tag {1}", + "image_path": "8b84131e0d8cdef3590d68b639033e6cb1f090876226e353ded7049fe0378c1f.jpg" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 55, + 451, + 296, + 475 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 451, + 296, + 475 + ], + "spans": [ + { + "bbox": [ + 55, + 451, + 296, + 475 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 55, + 451, + 296, + 475 + ], + "type": "inline_equation", + "content": "[x,y,z]" + }, + { + "bbox": [ + 55, + 451, + 296, + 475 + ], + "type": "text", + "content": " is the average value of all the coordinates of " + }, + { + "bbox": [ + 55, + 451, + 296, + 475 + ], + "type": "inline_equation", + "content": "V_{0}" + }, + { + "bbox": [ + 55, + 451, + 296, + 475 + ], + "type": "text", + "content": " in the dataset." + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 55, + 475, + 296, + 512 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 475, + 296, + 512 + ], + "spans": [ + { + "bbox": [ + 55, + 475, + 296, + 512 + ], + "type": "text", + "content": "In the time dimension, we randomly select 64 frames to standardize the video length and perform random mirroring and shifting of the video to generate augmented data." + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 55, + 521, + 239, + 535 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 521, + 239, + 535 + ], + "spans": [ + { + "bbox": [ + 55, + 521, + 239, + 535 + ], + "type": "text", + "content": "3.2. Self-Pacing Dropping Block (SPD)" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 55, + 541, + 295, + 624 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 541, + 295, + 624 + ], + "spans": [ + { + "bbox": [ + 55, + 541, + 295, + 624 + ], + "type": "text", + "content": "As shown in Fig. 3, our SPD consists of two components: \"Keypoint Activation\" and \"Weak Joints Dropping\". The input is a complete hand skeleton where the node information is independent, and the output is a high-dimensional feature skeleton after activation. The entire block is stacked three times, where the third block only outputs features and does not build a new skeleton." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 55, + 625, + 296, + 673 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 625, + 296, + 673 + ], + "spans": [ + { + "bbox": [ + 55, + 625, + 296, + 673 + ], + "type": "text", + "content": "Keypoint Activation Layer. In this part, we define the interaction of each joint with others as an activation. We employ a decoupled graph convolution [21] to perform this activation, represented as:" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 102, + 685, + 295, + 716 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 102, + 685, + 295, + 716 + ], + "spans": [ + { + "bbox": [ + 102, + 685, + 295, + 716 + ], + "type": "interline_equation", + "content": "F _ {o u t} ^ {t} = \\frac {1}{p} \\sum_ {k = 0} ^ {p} D ^ {- \\frac {1}{2}} A _ {k} D ^ {- \\frac {1}{2}} \\phi \\left(F _ {i n} ^ {t}\\right) \\tag {2}", + "image_path": "2d99d2ae58b8809d31bb2505fc87b4d590f15000039903f58298259ea1d1b864.jpg" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "spans": [ + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "inline_equation", + "content": "A" + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "content": " is the learnable adjacency matrix initialized as " + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "inline_equation", + "content": "I + In + Out" + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "content": ", " + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "inline_equation", + "content": "D" + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "content": " presents the diagonal degree of " + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "inline_equation", + "content": "A" + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "content": ". Here, " + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "inline_equation", + "content": "I" + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "content": " denotes self-connections, while " + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "inline_equation", + "content": "In" + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "inline_equation", + "content": "Out" + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "content": " represent inward and outward connections, respectively. The symbol " + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "inline_equation", + "content": "\\phi" + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "content": " represents a linear layer, and " + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "inline_equation", + "content": "p" + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "content": " represents the decoupling quantity, with a different " + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "inline_equation", + "content": "A" + }, + { + "bbox": [ + 313, + 247, + 555, + 356 + ], + "type": "text", + "content": " applied each time, followed by averaging the results. Following the graph convolution, we apply the simplest Temporal Convolutional Network (TCN) to capture features along the temporal dimension." + } + ] + } + ], + "index": 13 + }, + { + "bbox": [ + 313, + 358, + 556, + 456 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 358, + 556, + 456 + ], + "spans": [ + { + "bbox": [ + 313, + 358, + 556, + 456 + ], + "type": "text", + "content": "Weak Joints Dropping Layer. Compared to the simplified skeletons used in [21, 49], our approach utilizes a complete set of skeleton keypoints that contains richer information. However, this comes at the cost of significantly increasing the model's burden and raising the risk of introducing outlier points. To leverage these additional points with minimal overhead, we have designed an adaptive skeleton simplification strategy." + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 313, + 458, + 556, + 674 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 458, + 556, + 674 + ], + "spans": [ + { + "bbox": [ + 313, + 458, + 556, + 674 + ], + "type": "text", + "content": "Before formally introducing this layer, it is necessary to present our approach to skeleton decomposition. Based on the definition of VS described in Introduction, the decomposition concept involves grouping joints that exhibit strong overall correlation or frequently form VS together in sign language. These groupings can be viewed as a generalized or macro-level VS. A simple approach is grouping each finger and arm separately, but this creates too many groups, potentially affecting later interactions. Based on this, we conducted various experiments on VS, including treating individual fingers and distant fingers as VS. However, these experiments did not yield optimal results (the average performance in WLASL dataset decreased by " + }, + { + "bbox": [ + 313, + 458, + 556, + 674 + ], + "type": "inline_equation", + "content": "1.20\\%" + }, + { + "bbox": [ + 313, + 458, + 556, + 674 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 313, + 458, + 556, + 674 + ], + "type": "inline_equation", + "content": "1.64\\%" + }, + { + "bbox": [ + 313, + 458, + 556, + 674 + ], + "type": "text", + "content": ", respectively). We believe this is because the information from a single finger is insufficient to form a meaningful VS, while the information disparity between distant fingers is even greater. This also serves as the skeleton grouping basis for Figs. 4 and 5." + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "text", + "content": "After graph convolution in Eq. 2, " + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "inline_equation", + "content": "A(i,j)" + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "text", + "content": " has learned the weighted edge between node " + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "inline_equation", + "content": "i" + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "text", + "content": " and node " + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "inline_equation", + "content": "j" + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "text", + "content": ". These edges reflect the connection strength between " + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "inline_equation", + "content": "i" + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "inline_equation", + "content": "j" + }, + { + "bbox": [ + 313, + 677, + 556, + 715 + ], + "type": "text", + "content": ". Since these" + } + ] + } + ], + "index": 16 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "24323" + } + ] + } + ], + "index": 17 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 3 + }, + { + "para_blocks": [ + { + "bbox": [ + 55, + 72, + 296, + 107 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 72, + 296, + 107 + ], + "spans": [ + { + "bbox": [ + 55, + 72, + 296, + 107 + ], + "type": "text", + "content": "connections are not normalized via softmax, we can use " + }, + { + "bbox": [ + 55, + 72, + 296, + 107 + ], + "type": "inline_equation", + "content": "A" + }, + { + "bbox": [ + 55, + 72, + 296, + 107 + ], + "type": "text", + "content": " as a guide to calculate the total correlation of all keypoints, Cor, as follows:" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 94, + 118, + 295, + 143 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 94, + 118, + 295, + 143 + ], + "spans": [ + { + "bbox": [ + 94, + 118, + 295, + 143 + ], + "type": "interline_equation", + "content": "C o r _ {i} = \\sum_ {j} \\max \\left(A _ {0} (i, j), A _ {1}, \\dots , A _ {p}\\right) \\tag {3}", + "image_path": "af8d76436ac046b758e26e42a7655bbd2fb6ff9e96c596274081c4cf9cab4093.jpg" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 55, + 146, + 296, + 218 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 146, + 296, + 218 + ], + "spans": [ + { + "bbox": [ + 55, + 146, + 296, + 218 + ], + "type": "text", + "content": "A lower " + }, + { + "bbox": [ + 55, + 146, + 296, + 218 + ], + "type": "inline_equation", + "content": "Cor" + }, + { + "bbox": [ + 55, + 146, + 296, + 218 + ], + "type": "text", + "content": " to other keypoints indicates that the point is more independent. We refer to the two most independent points in each group as weak joints, which are dropped from the skeleton. To ensure that the number of keypoints in each group remains consistent, the groups involving both arms will have two fewer joints dropped." + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 224, + 126, + 236 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 224, + 126, + 236 + ], + "spans": [ + { + "bbox": [ + 55, + 224, + 126, + 236 + ], + "type": "text", + "content": "3.3.VSformer" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 243, + 296, + 326 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 243, + 296, + 326 + ], + "spans": [ + { + "bbox": [ + 55, + 243, + 296, + 326 + ], + "type": "text", + "content": "After the joint fusion block, we obtained six different joint groupings, as shown in Fig. 4(c). Based on the concept of skeletal disassembly, we treat each grouping as a generalized VS. We now need to focus on two main questions: Q1, what are the relationships between different VS? And Q2, do these VS remain consistent throughout the entire video, or do they change after a brief period?" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 327, + 296, + 421 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 327, + 296, + 421 + ], + "spans": [ + { + "bbox": [ + 55, + 327, + 296, + 421 + ], + "type": "text", + "content": "Inspired by [11], which proposes an efficient model for partitioning human limbs in action recognition, we employ two types of self-attention to constrain the model's focus on the aforementioned questions. Specifically, we calculate spatial pairing attention among different combinations within a frame to address Q1, and we compute temporal window attention for both global and local motion of the combinations across different frames to address Q2." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 422, + 296, + 540 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 422, + 296, + 540 + ], + "spans": [ + { + "bbox": [ + 55, + 422, + 296, + 540 + ], + "type": "text", + "content": "Spatial Pairing Attention (SPA). Given a frame " + }, + { + "bbox": [ + 55, + 422, + 296, + 540 + ], + "type": "inline_equation", + "content": "F_{t} \\in \\mathbb{R}^{G \\times C}" + }, + { + "bbox": [ + 55, + 422, + 296, + 540 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 55, + 422, + 296, + 540 + ], + "type": "inline_equation", + "content": "G = 6" + }, + { + "bbox": [ + 55, + 422, + 296, + 540 + ], + "type": "text", + "content": " represents VS groupings, the ideal approach would involve calculating attention among all VS pairwise. However, this is inefficient and often purposeless, as VS that are far apart in the skeletal structure tend to have weak correlations. For instance, in the sign language gesture for \"accent\", it is unnecessary to compare the fingers of the left hand with those of the unused right hand. Based on this observation, we ultimately partition " + }, + { + "bbox": [ + 55, + 422, + 296, + 540 + ], + "type": "inline_equation", + "content": "F_{t}" + }, + { + "bbox": [ + 55, + 422, + 296, + 540 + ], + "type": "text", + "content": " into pairs as follows:" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 128, + 541, + 294, + 555 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 128, + 541, + 294, + 555 + ], + "spans": [ + { + "bbox": [ + 128, + 541, + 294, + 555 + ], + "type": "interline_equation", + "content": "F _ {t} = \\left[ P _ {\\text {a r m}}, P _ {\\text {l e f t}}, P _ {\\text {r i g h t}} \\right] \\tag {4}", + "image_path": "489b009c072f68db68617aa8cd4831bc24a112bbd8f0ce60f19232d9b94826bd.jpg" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 55, + 558, + 296, + 605 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 558, + 296, + 605 + ], + "spans": [ + { + "bbox": [ + 55, + 558, + 296, + 605 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 55, + 558, + 296, + 605 + ], + "type": "inline_equation", + "content": "P_{\\mathrm{arm}} \\in \\mathbb{R}^{2 \\times C}" + }, + { + "bbox": [ + 55, + 558, + 296, + 605 + ], + "type": "text", + "content": " represents the grouping of both arms, " + }, + { + "bbox": [ + 55, + 558, + 296, + 605 + ], + "type": "inline_equation", + "content": "P_{\\mathrm{left}}" + }, + { + "bbox": [ + 55, + 558, + 296, + 605 + ], + "type": "text", + "content": " denotes the four fingers of the left hand, and " + }, + { + "bbox": [ + 55, + 558, + 296, + 605 + ], + "type": "inline_equation", + "content": "P_{\\mathrm{right}}" + }, + { + "bbox": [ + 55, + 558, + 296, + 605 + ], + "type": "text", + "content": " represents the remaining four fingers of the right hand. Represented as:" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 86, + 609, + 295, + 637 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 86, + 609, + 295, + 637 + ], + "spans": [ + { + "bbox": [ + 86, + 609, + 295, + 637 + ], + "type": "interline_equation", + "content": "S P A _ {\\text {o u t}} = \\operatorname {c o n c a t} \\left(\\operatorname {s o f t m a x} \\left(\\frac {Q _ {i} K _ {i} ^ {T}}{\\sqrt {h}}\\right) V _ {i}\\right) \\tag {5}", + "image_path": "65832fd31c530aa01aa71498430084165004ef934a9b2f3db123a9ddd11e8ce1.jpg" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 55, + 641, + 295, + 689 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 641, + 295, + 689 + ], + "spans": [ + { + "bbox": [ + 55, + 641, + 295, + 689 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 55, + 641, + 295, + 689 + ], + "type": "inline_equation", + "content": "Q_{i}, K_{i}, V_{i}" + }, + { + "bbox": [ + 55, + 641, + 295, + 689 + ], + "type": "text", + "content": " are obtained from " + }, + { + "bbox": [ + 55, + 641, + 295, + 689 + ], + "type": "inline_equation", + "content": "P_{i}" + }, + { + "bbox": [ + 55, + 641, + 295, + 689 + ], + "type": "text", + "content": " after mapping through linear layers. The attention is actually calculated among the 8 joints within each pairing and is ultimately concatenated to form the output." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 55, + 689, + 296, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 689, + 296, + 713 + ], + "spans": [ + { + "bbox": [ + 55, + 689, + 296, + 713 + ], + "type": "text", + "content": "Temporal Window Attention (TWA). Depending on different sign language videos, the VS can be classified into" + } + ] + } + ], + "index": 11 + }, + { + "type": "image", + "bbox": [ + 321, + 76, + 553, + 107 + ], + "blocks": [ + { + "bbox": [ + 321, + 76, + 553, + 107 + ], + "lines": [ + { + "bbox": [ + 321, + 76, + 553, + 107 + ], + "spans": [ + { + "bbox": [ + 321, + 76, + 553, + 107 + ], + "type": "image", + "image_path": "621111fb7906cca097896dddac1e940b4f52be4a0b9723b1c4add3959b904881.jpg" + } + ] + } + ], + "index": 12, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 313, + 121, + 555, + 143 + ], + "lines": [ + { + "bbox": [ + 313, + 121, + 555, + 143 + ], + "spans": [ + { + "bbox": [ + 313, + 121, + 555, + 143 + ], + "type": "text", + "content": "Figure 6. TWA windows. The small ticks represent frames, and the large ticks represent windows." + } + ] + } + ], + "index": 13, + "angle": 0, + "type": "image_caption" + } + ], + "index": 12 + }, + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "spans": [ + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "text", + "content": "two types: short duration and long duration, representing their respective duration periods. Accordingly, our TWA is also divided into two types: TWA-s and TWA-l. Given a video " + }, + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "inline_equation", + "content": "X \\in \\mathbb{R}^{T \\times G \\times C}" + }, + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "inline_equation", + "content": "T" + }, + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "text", + "content": " represents the total number of frames, the timeline is divided into " + }, + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "text", + "content": " windows, each containing " + }, + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "inline_equation", + "content": "k" + }, + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "text", + "content": " frames, such that " + }, + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "inline_equation", + "content": "T = n \\times k" + }, + { + "bbox": [ + 313, + 165, + 555, + 297 + ], + "type": "text", + "content": ", as shown in Fig. 6. For TWA-s, attention is computed within each window. For TWA-l, frames with the same index are selected sequentially from each window. Finally, the results from both types of attention are concatenated to produce the output, represented as:" + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 328, + 305, + 555, + 320 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 328, + 305, + 555, + 320 + ], + "spans": [ + { + "bbox": [ + 328, + 305, + 555, + 320 + ], + "type": "interline_equation", + "content": "T W A _ {o u t} = \\operatorname {s o f t m a x} \\left(\\phi_ {1} (W) \\phi_ {2} ^ {T} (W) + b\\right) \\phi_ {3} (W) \\tag {6}", + "image_path": "d5378820620685e8d4631189e7aac3e620e089adfef440e8d2289297af58beaf.jpg" + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "spans": [ + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "inline_equation", + "content": "W" + }, + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "text", + "content": " represents the selected windows, " + }, + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "inline_equation", + "content": "\\phi_{n}" + }, + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "text", + "content": " denotes dividing the " + }, + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "inline_equation", + "content": "C" + }, + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "text", + "content": " dimension into three equal parts and selecting the " + }, + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "text", + "content": "-th part, and " + }, + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "inline_equation", + "content": "b" + }, + { + "bbox": [ + 313, + 328, + 554, + 363 + ], + "type": "text", + "content": " represents the added bias." + } + ] + } + ], + "index": 16 + }, + { + "bbox": [ + 313, + 365, + 555, + 533 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 365, + 555, + 533 + ], + "spans": [ + { + "bbox": [ + 313, + 365, + 555, + 533 + ], + "type": "text", + "content": "It is worth noting that the discussions on Q1 and Q2 are not fully addressed by these two attention layers. The SPA only addresses the relationships between paired VS, while the windows defined in TWA do not entirely correspond to the duration of the VS. Therefore, we incorporate a simple GCN and TCN to enhance the model's ability to address the remaining scenarios. The connection method of SPA and TWA is shown in the right side of Fig. 3. After passing the input data " + }, + { + "bbox": [ + 313, + 365, + 555, + 533 + ], + "type": "inline_equation", + "content": "X \\in \\mathbb{R}^{T \\times G \\times C}" + }, + { + "bbox": [ + 313, + 365, + 555, + 533 + ], + "type": "text", + "content": " through a linear layer, we divide its channel dimension " + }, + { + "bbox": [ + 313, + 365, + 555, + 533 + ], + "type": "inline_equation", + "content": "C" + }, + { + "bbox": [ + 313, + 365, + 555, + 533 + ], + "type": "text", + "content": " into four equal parts, sending each part into one of the four modules and later concatenated. Additionally, unlike the Decoupled GCN in Sect. 3.2, the GCN here utilizes an adjacency matrix shared across certain frames, which significantly reduces the computational load." + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 313, + 539, + 462, + 552 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 539, + 462, + 552 + ], + "spans": [ + { + "bbox": [ + 313, + 539, + 462, + 552 + ], + "type": "text", + "content": "3.4. Multi-Grouping Ensemble" + } + ] + } + ], + "index": 18 + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "spans": [ + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "text", + "content": "In the previous sections, we introduced the concept of skeletal disassembly based on VS. However, since we cannot precisely determine the true VS composition for each sign language, using a fixed disassembly method appears to be a compromise, potentially weakening interpretability. Therefore, we adopt an ensemble strategy to reduce such discrepancies. Specifically, we designed two additional skeleton grouping schemes, as shown in the Fig. 5, and trained the model separately to obtain the output " + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "inline_equation", + "content": "q" + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "text", + "content": " from the final fully connected layer for each scheme. Following [21], we aggregate these results using a weighted sum, represented as: " + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "inline_equation", + "content": "q_{\\mathrm{total}} = \\alpha_1q_{\\mathrm{type - 1}} + \\alpha_2q_{\\mathrm{type - 2}} + \\alpha_3q_{\\mathrm{type - 3}} + \\alpha_4q_{\\mathrm{bone}}" + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "inline_equation", + "content": "\\alpha" + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "text", + "content": " represents the weights. For a detailed discussion on " + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "inline_equation", + "content": "\\alpha" + }, + { + "bbox": [ + 313, + 558, + 555, + 713 + ], + "type": "text", + "content": "," + } + ] + } + ], + "index": 19 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 318, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 757 + ], + "type": "text", + "content": "24324" + } + ] + } + ], + "index": 20 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 4 + }, + { + "para_blocks": [ + { + "type": "table", + "bbox": [ + 56, + 70, + 555, + 271 + ], + "blocks": [ + { + "bbox": [ + 56, + 70, + 555, + 271 + ], + "lines": [ + { + "bbox": [ + 56, + 70, + 555, + 271 + ], + "spans": [ + { + "bbox": [ + 56, + 70, + 555, + 271 + ], + "type": "table", + "html": "
MethodsMSASL1000MSASL200MSASL100
P-IP-CP-IP-CP-IP-C
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
RGB-based
HMA [15]49.1669.7546.2768.6066.3084.0367.4784.0373.4589.7074.5989.70
I3D [5]*--57.6981.05--81.9793.79--81.7695.16
BSL [2]64.7185.5961.5584.43--------
TCK [33]†----80.3191.8281.1492.2483.0493.4683.9193.52
skeleton-based
ST-GCN [58]36.0359.9232.3257.1552.9176.6754.2077.6259.8482.0360.7982.96
BEST [60]58.8281.1854.8780.0576.6091.5476.7591.9580.9895.1181.2495.44
SignBERT [14]†59.8081.8657.0680.9477.3491.1078.0291.4881.3793.6682.3193.76
SignBERT+ [17]†62.4283.4960.1582.4478.51-79.35-84.94-85.23-
MASA [61]†63.4783.8960.7983.2979.2592.8679.7093.3383.2295.2483.1995.46
SSRL [62]†65.2285.0962.6884.3882.9394.7883.4395.0386.2696.9686.6396.79
Ours70.7186.3668.3885.7184.6294.0485.1994.0886.2695.2486.6895.39
Ours(4-crops)73.1888.6470.8887.5986.1795.5186.6495.6188.1197.0988.3297.18
", + "image_path": "70acf9951ff5aaf6d8c7cd56e115a61e787330cd2da58e5baa16e5c647a0f337.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "table_body" + } + ], + "index": 0 + }, + { + "type": "table", + "bbox": [ + 56, + 312, + 555, + 531 + ], + "blocks": [ + { + "bbox": [ + 55, + 278, + 555, + 301 + ], + "lines": [ + { + "bbox": [ + 55, + 278, + 555, + 301 + ], + "spans": [ + { + "bbox": [ + 55, + 278, + 555, + 301 + ], + "type": "text", + "content": "Table 1. Comparison with previous work on MSASL. “†” indicates that the model was pre-trained, and “*” indicates methods that applied result fusion with other modalities in post-processing. Bold indicates the best performance, while underlined denotes the second-best." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "table_caption" + }, + { + "bbox": [ + 56, + 312, + 555, + 531 + ], + "lines": [ + { + "bbox": [ + 56, + 312, + 555, + 531 + ], + "spans": [ + { + "bbox": [ + 56, + 312, + 555, + 531 + ], + "type": "table", + "html": "
MethodsWLASL2000WLASL300WLASL100
P-IP-CP-IP-CP-IP-C
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
RGB-based
I3D [5]*32.4857.31--56.1479.9456.2478.3865.8984.1167.0184.58
HMA [15]37.9171.2635.9070.00--------
BSL [2]46.8279.3644.7278.47--------
TCK [33]†----68.5689.5268.7589.4177.5291.0877.5591.42
NLA-SLR V64 [63]*51.1583.4348.1482.20--------
skeleton-based
P3D [30]44.4779.6942.1878.5267.1889.0167.6289.2476.7191.9778.2792.97
BEST [60]46.2579.3343.5277.6567.6689.2268.3189.5777.9191.4777.8392.50
SignBERT [14]†47.4683.3245.1782.3270.3688.9271.1789.3679.0793.8080.0594.17
MASA [61]†49.0682.9046.9181.8073.6591.7774.3392.1383.7293.8084.4794.30
NLA-SLR K64 [63]*49.1082.0046.1880.71--------
SL-GCN [22]*51.5084.9448.8784.02--------
DSTA [18]53.68-51.17-79.97-80.56-82.38-83.09-
Ours53.5483.1851.1882.0379.0494.0179.7494.2584.5094.5785.2594.33
Ours(4-crops)55.9887.0753.5486.0480.0994.7680.8594.9285.6694.9686.2595.08
", + "image_path": "948ee5102452f7798c718f2b06fb8bcf7dfdc78e35f33a87fe17a339fff5a94e.jpg" + } + ] + } + ], + "index": 2, + "angle": 0, + "type": "table_body" + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 539, + 555, + 562 + ], + "lines": [ + { + "bbox": [ + 55, + 539, + 555, + 562 + ], + "spans": [ + { + "bbox": [ + 55, + 539, + 555, + 562 + ], + "type": "text", + "content": "Table 2. Comparison with previous work on WLASL. “†” indicates that the model was pre-trained, and “*” indicates methods that applied result fusion with other modalities in post-processing. Bold indicates the best performance, while underlined denotes the second-best." + } + ] + } + ], + "index": 3, + "angle": 0, + "type": "text" + }, + { + "bbox": [ + 55, + 582, + 295, + 631 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 582, + 295, + 631 + ], + "spans": [ + { + "bbox": [ + 55, + 582, + 295, + 631 + ], + "type": "text", + "content": "please refer to Sect. 4.4. The term bone refers to the joint data of type-1 represented in vector form, generated by directing from the source joint to its target joint according to the natural connections of the human body." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 639, + 137, + 653 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 639, + 137, + 653 + ], + "spans": [ + { + "bbox": [ + 55, + 639, + 137, + 653 + ], + "type": "text", + "content": "4. Experiments" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 659, + 168, + 672 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 659, + 168, + 672 + ], + "spans": [ + { + "bbox": [ + 55, + 659, + 168, + 672 + ], + "type": "text", + "content": "4.1. Parameter Settings" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 55, + 677, + 296, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 677, + 296, + 715 + ], + "spans": [ + { + "bbox": [ + 55, + 677, + 296, + 715 + ], + "type": "text", + "content": "We employ the AdamW optimizer with a plateau learning rate schedule. The initial learning rate is set to " + }, + { + "bbox": [ + 55, + 677, + 296, + 715 + ], + "type": "inline_equation", + "content": "1 \\times 10^{-3}" + }, + { + "bbox": [ + 55, + 677, + 296, + 715 + ], + "type": "text", + "content": ", with a minimum learning rate of " + }, + { + "bbox": [ + 55, + 677, + 296, + 715 + ], + "type": "inline_equation", + "content": "1 \\times 10^{-5}" + }, + { + "bbox": [ + 55, + 677, + 296, + 715 + ], + "type": "text", + "content": " and a decay rate" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "spans": [ + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "text", + "content": "of 0.5. The batch size is 32 and the loss function used is cross-entropy with label smoothing. Training spans a maximum of 200 epochs, with the first 15 epochs serving as a warm-up period starting from a learning rate of " + }, + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "inline_equation", + "content": "1 \\times 10^{-7}" + }, + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "text", + "content": ". A random sample of 64 frames is selected as input, and with a " + }, + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "inline_equation", + "content": "50\\%" + }, + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "text", + "content": " probability, image mirroring is applied. A random jitter in the range " + }, + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "inline_equation", + "content": "(-10, 10)" + }, + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "text", + "content": " is added to each joint point. In the last two layers of SPD, joints are randomly masked with a probability of " + }, + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "inline_equation", + "content": "(1 - p_{kr}) / 100 \\times \\text{epoch}" + }, + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "inline_equation", + "content": "p_{kr}" + }, + { + "bbox": [ + 313, + 582, + 555, + 703 + ], + "type": "text", + "content": " is set to 0.9. All experiments were performed with NVIDIA A100." + } + ] + } + ], + "index": 8 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "24325" + } + ] + } + ], + "index": 9 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 5 + }, + { + "para_blocks": [ + { + "type": "table", + "bbox": [ + 99, + 70, + 254, + 229 + ], + "blocks": [ + { + "bbox": [ + 99, + 70, + 254, + 229 + ], + "lines": [ + { + "bbox": [ + 99, + 70, + 254, + 229 + ], + "spans": [ + { + "bbox": [ + 99, + 70, + 254, + 229 + ], + "type": "table", + "html": "
MethodsNMFs-CSR
Top-1Top-5
RGB-based
I3D [5]*64.488.0
HMA [15]64.791.0
GLE-Net [36]69.088.1
skeleton-based
ST-GCN [58]59.986.8
BEST [60]68.594.4
MASA [61]†71.797.0
SignBERT [14]†74.993.2
Ours75.395.4
Ours(4-crops)76.797.2
", + "image_path": "4b0b34d9395710f0472923dc42a5987e8754994ed5e4212f2acabb0bcd27597a.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "table_body" + } + ], + "index": 0 + }, + { + "type": "table", + "bbox": [ + 87, + 271, + 266, + 360 + ], + "blocks": [ + { + "bbox": [ + 55, + 237, + 296, + 260 + ], + "lines": [ + { + "bbox": [ + 55, + 237, + 296, + 260 + ], + "spans": [ + { + "bbox": [ + 55, + 237, + 296, + 260 + ], + "type": "text", + "content": "Table 3. Comparison on NMFs-CSR. Bold indicates the best performance, while underlined denotes the second-best." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "table_caption" + }, + { + "bbox": [ + 87, + 271, + 266, + 360 + ], + "lines": [ + { + "bbox": [ + 87, + 271, + 266, + 360 + ], + "spans": [ + { + "bbox": [ + 87, + 271, + 266, + 360 + ], + "type": "table", + "html": "
MethodsSlovo
Top-1Top-5
Swin-large-48 [38]55.66-
MViTv2-small-48 [35]62.18-
MViTv2-small-32 [35]64.09-
Ours76.5393.08
Ours(4-crops)77.8395.17
", + "image_path": "6e0863849cf0b9d8091a93d7d3cd76aaf00e4eafdbdb07ae6c335cc5f9120a23.jpg" + } + ] + } + ], + "index": 2, + "angle": 0, + "type": "table_body" + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 366, + 295, + 391 + ], + "lines": [ + { + "bbox": [ + 55, + 366, + 295, + 391 + ], + "spans": [ + { + "bbox": [ + 55, + 366, + 295, + 391 + ], + "type": "text", + "content": "Table 4. Comparison with the baselines on Slovo. The final number indicates the number of frames input to the model." + } + ] + } + ], + "index": 3, + "angle": 0, + "type": "text" + }, + { + "bbox": [ + 55, + 411, + 296, + 507 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 411, + 296, + 507 + ], + "spans": [ + { + "bbox": [ + 55, + 411, + 296, + 507 + ], + "type": "text", + "content": "Metrics. Following [22, 61, 63], we report top-1 and top-5 classification accuracy, as well as per-instance (P-I) and per-class (P-C) precision metrics. Each table includes results for single-crop and 4-crop inference, with all ablation studies conducted in the single-crop setting. For 4-crops inference, we sum the predictions from the three types of skeleton groupings along with the bone data to obtain the final prediction." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 515, + 115, + 526 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 515, + 115, + 526 + ], + "spans": [ + { + "bbox": [ + 55, + 515, + 115, + 526 + ], + "type": "text", + "content": "4.2. Dataset" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 533, + 295, + 581 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 533, + 295, + 581 + ], + "spans": [ + { + "bbox": [ + 55, + 533, + 295, + 581 + ], + "type": "text", + "content": "We conduct experiments on four publicly available benchmarks: WLASL, MSASL, NMFs-CSL, and Slovo. All training occurs on the training set, with hyperparameter tuning performed on the validation set." + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 55, + 582, + 295, + 653 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 582, + 295, + 653 + ], + "spans": [ + { + "bbox": [ + 55, + 582, + 295, + 653 + ], + "type": "text", + "content": "MSASL [26] is a widely used dataset for American Sign Language (ASL). It contains a total of 16,054, 5,287, and 4,172 samples in the training, validation, and test sets, respectively, with a vocabulary of 1,000 signs. Two subsets, consisting of the top 100 and 300 words, are defined as MSASL100 and MSASL300, respectively." + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 55, + 653, + 296, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 653, + 296, + 713 + ], + "spans": [ + { + "bbox": [ + 55, + 653, + 296, + 713 + ], + "type": "text", + "content": "WLASL [32] has a larger vocabulary but fewer total samples than MSASL. It contains 2,000 signs performed by over 100 signers, with 14,289, 3,916, and 2,878 samples in the training, validation, and test sets. Similarly, two subsets, WLASL100 and WLASL300, are created. Both" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 313, + 72, + 553, + 96 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 72, + 553, + 96 + ], + "spans": [ + { + "bbox": [ + 313, + 72, + 553, + 96 + ], + "type": "text", + "content": "ASL datasets are collected from real-world scenarios, offering greater authenticity." + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 97, + 553, + 144 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 97, + 553, + 144 + ], + "spans": [ + { + "bbox": [ + 313, + 97, + 553, + 144 + ], + "type": "text", + "content": "NMFs-CSR [16] is a challenging Chinese Sign Language (CSL) dataset with a vocabulary of 1,067 and consists of 25,608 and 6,402 samples in the training and test sets, respectively." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 313, + 145, + 555, + 205 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 145, + 555, + 205 + ], + "spans": [ + { + "bbox": [ + 313, + 145, + 555, + 205 + ], + "type": "text", + "content": "Slovo [27] is the latest dataset for Russian Sign Language (RSL). It contains 20,000 videos across 1,000 isolated RSL gestures from 194 signers. Notably, the authors provide 42 hand landmarks based on MediaPipe, so we do not re-extract joint points." + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 313, + 214, + 537, + 227 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 214, + 537, + 227 + ], + "spans": [ + { + "bbox": [ + 313, + 214, + 537, + 227 + ], + "type": "text", + "content": "4.3. Comparison with State-of-the-art Methods" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 313, + 233, + 553, + 293 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 233, + 553, + 293 + ], + "spans": [ + { + "bbox": [ + 313, + 233, + 553, + 293 + ], + "type": "text", + "content": "In this section, we compare our model with several state-of-the-art approaches. For single-modality methods, we categorize them as either RGB-based or skeleton-based. For multi-modality methods, we report the performance of their single-modality models before ensemble." + } + ] + } + ], + "index": 13 + }, + { + "bbox": [ + 313, + 293, + 555, + 497 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 293, + 555, + 497 + ], + "spans": [ + { + "bbox": [ + 313, + 293, + 555, + 497 + ], + "type": "text", + "content": "MSASL [26]. As shown in the Tab. 1, RGB-based methods have developed slowly in recent years, and our approach shows a significant improvement when compared to them. Among skeleton-based methods, SignBERT [14] and BEST [60] designed self-supervised learning strategies to pre-train the model in order to alleviate issues related to limited interpretability and overfitting due to the small size of the SLR datasets. NASA [61] and SSRL [62] address the problem of neglecting explicit motion information and lexical signs during pre-training. However, none of these methods take into account the linguistic characteristics of sign language itself. Compared to them, our method achieves a substantial top-1 improvement of " + }, + { + "bbox": [ + 313, + 293, + 555, + 497 + ], + "type": "inline_equation", + "content": "7.96\\%" + }, + { + "bbox": [ + 313, + 293, + 555, + 497 + ], + "type": "text", + "content": " in P-I and " + }, + { + "bbox": [ + 313, + 293, + 555, + 497 + ], + "type": "inline_equation", + "content": "8.2\\%" + }, + { + "bbox": [ + 313, + 293, + 555, + 497 + ], + "type": "text", + "content": " in P-C on the full dataset. However, SSRL performs slightly better than our single-crop method on the 100-class subset, which we speculate is due to SPD requiring a certain amount of data to effectively learn the droppable nodes." + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 313, + 498, + 554, + 640 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 498, + 554, + 640 + ], + "spans": [ + { + "bbox": [ + 313, + 498, + 554, + 640 + ], + "type": "text", + "content": "WLASL [32]. SAM-SLR [22] and NLA-SLR [63] are currently the two best-performing heavy multi-modality ensemble models. Among them, SL-GCN is the skeleton model used in SAM-SLR for later ensemble, employing a multi-stream approach to further improve performance. K64 and V64 are skeleton and RGB models in NLASLR that use 64-frame input, where the skeleton is represented as a heatmap sequence, allowing their keypoint encoder to share the same architecture with the video encoder. DSTA [18] was the previously most accurate model. Compared to their results, our top-1 accuracy achieved a " + }, + { + "bbox": [ + 313, + 498, + 554, + 640 + ], + "type": "inline_equation", + "content": "2.30\\%" + }, + { + "bbox": [ + 313, + 498, + 554, + 640 + ], + "type": "text", + "content": " improvement on WLASL2000." + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 313, + 641, + 554, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 641, + 554, + 713 + ], + "spans": [ + { + "bbox": [ + 313, + 641, + 554, + 713 + ], + "type": "text", + "content": "NMFs-CSL [16]. As shown in Tab. 3, NMFs-CSL is a balanced dataset in terms of class distribution, so we only report the P-I accuracy. The pretraining strategy of SignBERT [14] is highly effective on this dataset; however, its performance on top-5 is suboptimal. Our model does not use any additional training data, yet it shows a significant" + } + ] + } + ], + "index": 16 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "text", + "content": "24326" + } + ] + } + ], + "index": 17 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 6 + }, + { + "para_blocks": [ + { + "type": "table", + "bbox": [ + 63, + 70, + 547, + 137 + ], + "blocks": [ + { + "bbox": [ + 63, + 70, + 547, + 137 + ], + "lines": [ + { + "bbox": [ + 63, + 70, + 547, + 137 + ], + "spans": [ + { + "bbox": [ + 63, + 70, + 547, + 137 + ], + "type": "table", + "html": "
SkeletonMSASL1000MSASL200MSASL100WLASL2000WLASL300WLASL100NMFs-CLR
Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5Top-1Top-5
type-170.7186.3684.6294.0486.2695.2453.5483.1879.0494.0184.5094.9675.2695.47
bone68.4185.8882.4193.3884.6894.0650.7682.4574.7091.1784.1193.8073.6394.31
type-270.3586.1085.2894.1187.4595.6453.0283.3278.4493.4183.7293.8074.7094.86
type-370.3586.1084.9294.1185.6095.1152.2682.4978.1493.4184.8893.4174.8795.55
", + "image_path": "b906ec28865617bbba0152fb78ccecece5a76f63ed3253cd0e0fe93ef268ec63.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "table_body" + } + ], + "index": 0 + }, + { + "type": "table", + "bbox": [ + 58, + 175, + 292, + 220 + ], + "blocks": [ + { + "bbox": [ + 180, + 145, + 428, + 156 + ], + "lines": [ + { + "bbox": [ + 180, + 145, + 428, + 156 + ], + "spans": [ + { + "bbox": [ + 180, + 145, + 428, + 156 + ], + "type": "text", + "content": "Table 5. Performance of multi-grouping VSformer on three datasets." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "table_caption" + }, + { + "bbox": [ + 58, + 175, + 292, + 220 + ], + "lines": [ + { + "bbox": [ + 58, + 175, + 292, + 220 + ], + "spans": [ + { + "bbox": [ + 58, + 175, + 292, + 220 + ], + "type": "table", + "html": "
weightingWLASL2000MSASL1000NMFs-CSL
Top-1Top-5Top-1Top-5Top-1Top-5
sum55.9887.0773.1888.6476.6597.17
acc56.0187.1873.1388.6476.6597.17
", + "image_path": "31e604748199a8431fe8b374c23dd2cd44acb63279e6389e154223e2aed8487b.jpg" + } + ] + } + ], + "index": 2, + "angle": 0, + "type": "table_body" + } + ], + "index": 2 + }, + { + "type": "table", + "bbox": [ + 76, + 251, + 275, + 314 + ], + "blocks": [ + { + "bbox": [ + 82, + 229, + 269, + 239 + ], + "lines": [ + { + "bbox": [ + 82, + 229, + 269, + 239 + ], + "spans": [ + { + "bbox": [ + 82, + 229, + 269, + 239 + ], + "type": "text", + "content": "Table 6. Performance of the two ensemble methods" + } + ] + } + ], + "index": 3, + "angle": 0, + "type": "table_caption" + }, + { + "bbox": [ + 76, + 251, + 275, + 314 + ], + "lines": [ + { + "bbox": [ + 76, + 251, + 275, + 314 + ], + "spans": [ + { + "bbox": [ + 76, + 251, + 275, + 314 + ], + "type": "table", + "html": "
MethodsWLASL300WLASL2000
Top-1Top-5Top-1Top-5
Linear74.7092.8146.3277.24
GCN76.3591.7751.1880.82
SPD(ours)79.0494.0153.5483.18
", + "image_path": "a7028c8dd42754b2060fb0519da73a576752b18f71243ebf83ae6ff3357910fb.jpg" + } + ] + } + ], + "index": 4, + "angle": 0, + "type": "table_body" + } + ], + "index": 4 + }, + { + "type": "table", + "bbox": [ + 72, + 345, + 280, + 408 + ], + "blocks": [ + { + "bbox": [ + 70, + 323, + 280, + 334 + ], + "lines": [ + { + "bbox": [ + 70, + 323, + 280, + 334 + ], + "spans": [ + { + "bbox": [ + 70, + 323, + 280, + 334 + ], + "type": "text", + "content": "Table 7. Comparing our SPD with GCN and linear layers." + } + ] + } + ], + "index": 5, + "angle": 0, + "type": "table_caption" + }, + { + "bbox": [ + 72, + 345, + 280, + 408 + ], + "lines": [ + { + "bbox": [ + 72, + 345, + 280, + 408 + ], + "spans": [ + { + "bbox": [ + 72, + 345, + 280, + 408 + ], + "type": "table", + "html": "
SPATMAWLASL300WLASL2000
Top-1Top-5Top-1Top-5
✓77.6993.4152.4382.24
✓77.9993.2652.3681.83
✓✓79.0494.0153.5483.18
", + "image_path": "a41391df11b191da55a27851e8377608ec051734a9a098a54cd740bb7ff4d18e.jpg" + } + ] + } + ], + "index": 6, + "angle": 0, + "type": "table_body" + } + ], + "index": 6 + }, + { + "bbox": [ + 55, + 416, + 295, + 439 + ], + "lines": [ + { + "bbox": [ + 55, + 416, + 295, + 439 + ], + "spans": [ + { + "bbox": [ + 55, + 416, + 295, + 439 + ], + "type": "text", + "content": "Table 8. Ablation study on the impact of our two types of attention mechanisms." + } + ] + } + ], + "index": 7, + "angle": 0, + "type": "text" + }, + { + "bbox": [ + 55, + 460, + 295, + 495 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 460, + 295, + 495 + ], + "spans": [ + { + "bbox": [ + 55, + 460, + 295, + 495 + ], + "type": "text", + "content": "improvement over SignBERT in overall performance, with a " + }, + { + "bbox": [ + 55, + 460, + 295, + 495 + ], + "type": "inline_equation", + "content": "1.8\\%" + }, + { + "bbox": [ + 55, + 460, + 295, + 495 + ], + "type": "text", + "content": " improvement in top-1 accuracy and a " + }, + { + "bbox": [ + 55, + 460, + 295, + 495 + ], + "type": "inline_equation", + "content": "4\\%" + }, + { + "bbox": [ + 55, + 460, + 295, + 495 + ], + "type": "text", + "content": " improvement in top-5 accuracy." + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 55, + 496, + 295, + 604 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 496, + 295, + 604 + ], + "spans": [ + { + "bbox": [ + 55, + 496, + 295, + 604 + ], + "type": "text", + "content": "Slovo [27]. Slovo is a recently released RSL dataset, with only limited experimental data available. We provide an up-to-date benchmark for better evaluating this dataset. As shown in Tab. 4, the results for Swin [38] and MViTv2 [35] were provided as baselines by the dataset authors. Notably, the skeleton landmarks provided by the authors includes only 42 hand joints, without arm data, which partially limits our model's performance. Despite this, we still achieve results significantly above the baselines." + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 55, + 611, + 149, + 624 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 611, + 149, + 624 + ], + "spans": [ + { + "bbox": [ + 55, + 611, + 149, + 624 + ], + "type": "text", + "content": "4.4. Ablation Study" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 55, + 629, + 295, + 666 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 629, + 295, + 666 + ], + "spans": [ + { + "bbox": [ + 55, + 629, + 295, + 666 + ], + "type": "text", + "content": "In this section, we conduct several key ablation experiments on the WLASL dataset [32] and report P-I accuracy to validate the effectiveness of our approach." + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 55, + 665, + 296, + 714 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 665, + 296, + 714 + ], + "spans": [ + { + "bbox": [ + 55, + 665, + 296, + 714 + ], + "type": "text", + "content": "SPD. We first replace SPD with two alternative methods. As shown in Tab. 7, \"Linear\" denotes a three-layer Conv-2D module with the same output dimension as SPD, while \"GCN\" refers to three decoupled graph convolution" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 313, + 178, + 553, + 249 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 178, + 553, + 249 + ], + "spans": [ + { + "bbox": [ + 313, + 178, + 553, + 249 + ], + "type": "text", + "content": "layer without the weak joints dropping strategy. To ensure consistent output data shape, these two methods directly input the simplified 24 joint points. The complete SFB implementation achieves the highest accuracy, indicating that our dropping strategy better facilitates the extraction of sign language action features." + } + ] + } + ], + "index": 13 + }, + { + "bbox": [ + 313, + 251, + 554, + 358 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 251, + 554, + 358 + ], + "spans": [ + { + "bbox": [ + 313, + 251, + 554, + 358 + ], + "type": "text", + "content": "Multi-Grouping. As shown in Tab. 5, we present the final performance of four types of crops. The overall performance of each grouping type varies only slightly, mainly depending on the sample distribution of the dataset. However, integrating all four crops significantly enhances action classification performance. The performance of the bone-based grouping is slightly lower than that of joint-based results, primarily because our model is designed for points and is less sensitive to vector-based data." + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 313, + 361, + 554, + 469 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 361, + 554, + 469 + ], + "spans": [ + { + "bbox": [ + 313, + 361, + 554, + 469 + ], + "type": "text", + "content": "SAM-SLR [21] ultimately using manually set weights to control the ensemble. However, we argue that such weights lack interpretability. As shown in Tab. 6, we employed two methods to integrate the results: \"acc\" refers to using the individual top-1 accuracy of each crop as the weight, while \"sum\" represents directly summing the outputs of the final layer of the model. Both methods resulted in an about " + }, + { + "bbox": [ + 313, + 361, + 554, + 469 + ], + "type": "inline_equation", + "content": "2\\%" + }, + { + "bbox": [ + 313, + 361, + 554, + 469 + ], + "type": "text", + "content": " improvement in top-1 accuracy, demonstrating that our ensemble approach is both general and robust." + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 313, + 471, + 554, + 544 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 471, + 554, + 544 + ], + "spans": [ + { + "bbox": [ + 313, + 471, + 554, + 544 + ], + "type": "text", + "content": "VSformer. For our two types of self-attention, we replace one with a linear layer to test the performance of the other individually. As shown in Tab. 8, the full VSformer achieves significantly better performance, indicating that features in both temporal and spatial dimensions are crucial for distinguishing sign language actions." + } + ] + } + ], + "index": 16 + }, + { + "bbox": [ + 313, + 559, + 388, + 572 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 559, + 388, + 572 + ], + "spans": [ + { + "bbox": [ + 313, + 559, + 388, + 572 + ], + "type": "text", + "content": "5. Conclusion" + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 313, + 582, + 554, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 582, + 554, + 713 + ], + "spans": [ + { + "bbox": [ + 313, + 582, + 554, + 713 + ], + "type": "text", + "content": "In this work, we design a skeleton recognition model tailored specifically for ISLR, termed VSNet. The core idea of our approach is to focus on the truly essential elements of sign language. To this end, we first introduce a weak joints dropping strategy to prioritize more critical points in the skeleton. We then propose the VSformer to capture the linguistic characteristics of sign language, where Visual Symbols (VS) represent smaller components of sign language actions. We conduct extensive experiments to validate the effectiveness of the proposed VSNet, achieving state-of-the-art performance across four benchmarks." + } + ] + } + ], + "index": 18 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "24327" + } + ] + } + ], + "index": 19 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 7 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 72, + 158, + 85 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 72, + 158, + 85 + ], + "spans": [ + { + "bbox": [ + 56, + 72, + 158, + 85 + ], + "type": "text", + "content": "Acknowledgements" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 92, + 297, + 212 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 92, + 297, + 212 + ], + "spans": [ + { + "bbox": [ + 55, + 92, + 297, + 212 + ], + "type": "text", + "content": "This work was supported in part by National Natural Science Foundation of China (No. 62476052), Sichuan Science and Technology Program (No. 2024NSFSC1473), the open project of Sichuan Provincial Key Laboratory of Philosophy and Social Science for Language Intelligence in Special Education (No. YYZN-2023-3), the Humanities and Social Sciences Project of the Ministry of Education (No. 23YJA740013), and Central Guidance for Local Science and Technology Development Fund Projects (No. 2024ZYD0268)." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 224, + 115, + 237 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 224, + 115, + 237 + ], + "spans": [ + { + "bbox": [ + 56, + 224, + 115, + 237 + ], + "type": "text", + "content": "References" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 57, + 245, + 294, + 714 + ], + "type": "list", + "angle": 0, + "index": 14, + "blocks": [ + { + "bbox": [ + 61, + 245, + 294, + 310 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 61, + 245, + 294, + 310 + ], + "spans": [ + { + "bbox": [ + 61, + 245, + 294, + 310 + ], + "type": "text", + "content": "[1] Nikolas Adaloglou, Theocharis Chatzis, Ilias Papastratis, Andreas Stergioulas, Georgios Th Papadopoulos, Vassia Zacharopoulou, George J Xydopoulos, Klimnis Atzakas, Dimitris Papazachariou, and Petros Daras. A comprehensive study on deep learning-based methods for sign language recognition. IEEE TMM, 24:1750-1762, 2021. 1, 2" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 62, + 312, + 294, + 357 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 312, + 294, + 357 + ], + "spans": [ + { + "bbox": [ + 62, + 312, + 294, + 357 + ], + "type": "text", + "content": "[2] Samuel Albanie, Gül Varol, Liliane Momeni, Triantafyllos Afouras, Joon Son Chung, Neil Fox, and Andrew Zisserman. Bsl-1k: Scaling up co-articulated sign language recognition using mouthing cues. In ECCV. Springer, 2020. 6" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 62, + 358, + 294, + 392 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 358, + 294, + 392 + ], + "spans": [ + { + "bbox": [ + 62, + 358, + 294, + 392 + ], + "type": "text", + "content": "[3] Gedas Bertasius, Heng Wang, and Lorenzo Torresani. Is space-time attention all you need for video understanding? In ICML, page 4, 2021. 3" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 62, + 393, + 294, + 435 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 393, + 294, + 435 + ], + "spans": [ + { + "bbox": [ + 62, + 393, + 294, + 435 + ], + "type": "text", + "content": "[4] Amanda Cardoso Duarte, Samuel Albanie, Xavier Giro Nieto, and Gül Varol. Sign language video retrieval with freeform textual queries. In CVPR, pages 14074-14084, 2022. 2" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 62, + 438, + 294, + 472 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 438, + 294, + 472 + ], + "spans": [ + { + "bbox": [ + 62, + 438, + 294, + 472 + ], + "type": "text", + "content": "[5] Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. In CVPR, pages 6299-6308, 2017. 2, 6, 7" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 62, + 474, + 294, + 507 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 474, + 294, + 507 + ], + "spans": [ + { + "bbox": [ + 62, + 474, + 294, + 507 + ], + "type": "text", + "content": "[6] Joao Carreira and Andrew Zisserman. Quo vadis, action recognition? a new model and the kinetics dataset. In CVPR, pages 6299-6308, 2017. 3" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 62, + 509, + 294, + 552 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 509, + 294, + 552 + ], + "spans": [ + { + "bbox": [ + 62, + 509, + 294, + 552 + ], + "type": "text", + "content": "[7] Yutong Chen, Fangyun Wei, Xiao Sun, Zhirong Wu, and Stephen Lin. A simple multi-modality transfer learning baseline for sign language translation. In CVPR, pages 5120-5130, 2022. 1" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 62, + 554, + 294, + 598 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 554, + 294, + 598 + ], + "spans": [ + { + "bbox": [ + 62, + 554, + 294, + 598 + ], + "type": "text", + "content": "[8] Ke Cheng, Yifan Zhang, Congqi Cao, Lei Shi, Jian Cheng, and Hanqing Lu. Decoupling gcn with dropgraph module for skeleton-based action recognition. In ECCV, pages 536-553. Springer, 2020. 2" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 62, + 600, + 294, + 643 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 600, + 294, + 643 + ], + "spans": [ + { + "bbox": [ + 62, + 600, + 294, + 643 + ], + "type": "text", + "content": "[9] Xiangxiang Chu, Zhi Tian, Yuqing Wang, Bo Zhang, Haibing Ren, Xiaolin Wei, Huaxia Xia, and Chunhua Shen. Twins: Revisiting the design of spatial attention in vision transformers. NeurIPS, 34:9355-9366, 2021. 3" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 57, + 646, + 294, + 678 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 646, + 294, + 678 + ], + "spans": [ + { + "bbox": [ + 57, + 646, + 294, + 678 + ], + "type": "text", + "content": "[10] MMPose Contributors. Openmmlab pose estimation toolbox and benchmark. https://github.com/open-mmlab/mmpose, 2020.3" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 57, + 680, + 294, + 714 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 680, + 294, + 714 + ], + "spans": [ + { + "bbox": [ + 57, + 680, + 294, + 714 + ], + "type": "text", + "content": "[11] Jeonghyeok Do and Munchurl Kim. Skateformer: Skeletal-temporal transformer for human action recognition. arXiv preprint arXiv:2403.09508, 2024. 3, 5" + } + ] + } + ], + "index": 13 + } + ], + "sub_type": "ref_text" + }, + { + "bbox": [ + 316, + 73, + 553, + 714 + ], + "type": "list", + "angle": 0, + "index": 31, + "blocks": [ + { + "bbox": [ + 316, + 73, + 553, + 106 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 73, + 553, + 106 + ], + "spans": [ + { + "bbox": [ + 316, + 73, + 553, + 106 + ], + "type": "text", + "content": "[12] Yong Du, Wei Wang, and Liang Wang. Hierarchical recurrent neural network for skeleton based action recognition. In CVPR, pages 1110-1118, 2015. 1, 3" + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 317, + 108, + 553, + 140 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 108, + 553, + 140 + ], + "spans": [ + { + "bbox": [ + 317, + 108, + 553, + 140 + ], + "type": "text", + "content": "[13] Karen Emmorey. Language, cognition, and the brain: Insights from sign language research. Psychology Press, 2001. 1" + } + ] + } + ], + "index": 16 + }, + { + "bbox": [ + 317, + 143, + 553, + 186 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 143, + 553, + 186 + ], + "spans": [ + { + "bbox": [ + 317, + 143, + 553, + 186 + ], + "type": "text", + "content": "[14] Hezhen Hu, Weichao Zhao, Wengang Zhou, Yuechen Wang, and Houqiang Li. Signbert: Pre-training of hand-model-aware representation for sign language recognition. In ICCV, pages 11087-11096, 2021. 6, 7" + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 317, + 188, + 553, + 220 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 188, + 553, + 220 + ], + "spans": [ + { + "bbox": [ + 317, + 188, + 553, + 220 + ], + "type": "text", + "content": "[15] Hezhen Hu, Wengang Zhou, and Houqiang Li. Hand-model-aware sign language recognition. In AAAI, pages 1558-1566, 2021. 2, 6, 7" + } + ] + } + ], + "index": 18 + }, + { + "bbox": [ + 317, + 223, + 553, + 277 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 223, + 553, + 277 + ], + "spans": [ + { + "bbox": [ + 317, + 223, + 553, + 277 + ], + "type": "text", + "content": "[16] Hezhen Hu, Wengang Zhou, Junfu Pu, and Houqiang Li. Global-local enhancement network for nmf-aware sign language recognition. ACM transactions on multimedia computing, communications, and applications (TOMM), 17(3): 1-19, 2021. 2, 7" + } + ] + } + ], + "index": 19 + }, + { + "bbox": [ + 316, + 279, + 553, + 323 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 279, + 553, + 323 + ], + "spans": [ + { + "bbox": [ + 316, + 279, + 553, + 323 + ], + "type": "text", + "content": "[17] Hezhen Hu, Weichao Zhao, Wengang Zhou, and Houqiang Li. Signbert+: Hand-model-aware self-supervised pretraining for sign language understanding. IEEE TPAMI, 45 (9):11221-11239, 2023. 6" + } + ] + } + ], + "index": 20 + }, + { + "bbox": [ + 316, + 325, + 553, + 357 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 325, + 553, + 357 + ], + "spans": [ + { + "bbox": [ + 316, + 325, + 553, + 357 + ], + "type": "text", + "content": "[18] Lianyu Hu, Liqing Gao, Zekang Liu, and Wei Feng. Dynamic spatial-temporal aggregation for skeleton-aware sign language recognition. In LREC/COLING, 2024. 6, 7" + } + ] + } + ], + "index": 21 + }, + { + "bbox": [ + 316, + 359, + 553, + 392 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 359, + 553, + 392 + ], + "spans": [ + { + "bbox": [ + 316, + 359, + 553, + 392 + ], + "type": "text", + "content": "[19] Jie Huang, Wengang Zhou, Houqiang Li, and Weiping Li. Attention-based 3d-cnns for large-vocabulary sign language recognition. IEEE TCSVT, 29(9):2822-2832, 2018. 2" + } + ] + } + ], + "index": 22 + }, + { + "bbox": [ + 316, + 394, + 553, + 426 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 394, + 553, + 426 + ], + "spans": [ + { + "bbox": [ + 316, + 394, + 553, + 426 + ], + "type": "text", + "content": "[20] Jie Huang, Wengang Zhou, Qilin Zhang, Houqiang Li, and Weiping Li. Video-based sign language recognition without temporal segmentation. In AAAI, 2018. 2" + } + ] + } + ], + "index": 23 + }, + { + "bbox": [ + 316, + 429, + 553, + 461 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 429, + 553, + 461 + ], + "spans": [ + { + "bbox": [ + 316, + 429, + 553, + 461 + ], + "type": "text", + "content": "[21] Songyao Jiang, Bin Sun, Lichen Wang, Yue Bai, Kunpeng Li, and Yun Fu. Skeleton aware multi-modal sign language recognition. In CVPR, pages 3413-3423, 2021. 1, 3, 4, 5, 8" + } + ] + } + ], + "index": 24 + }, + { + "bbox": [ + 317, + 464, + 553, + 506 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 464, + 553, + 506 + ], + "spans": [ + { + "bbox": [ + 317, + 464, + 553, + 506 + ], + "type": "text", + "content": "[22] Songyao Jiang, Bin Sun, Lichen Wang, Yue Bai, Kunpeng Li, and Yun Fu. Sign language recognition via skeleton-aware multi-model ensemble. arXiv preprint arXiv:2110.06161, 2021. 6, 7" + } + ] + } + ], + "index": 25 + }, + { + "bbox": [ + 316, + 509, + 553, + 542 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 509, + 553, + 542 + ], + "spans": [ + { + "bbox": [ + 316, + 509, + 553, + 542 + ], + "type": "text", + "content": "[23] Tao Jiang, Necati Cihan Camgoz, and Richard Bowden. Skeleton: Skeletal transformers for robust body-posed estimation. In CVPR, pages 3394–3402, 2021. 1, 3" + } + ] + } + ], + "index": 26 + }, + { + "bbox": [ + 317, + 544, + 553, + 587 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 544, + 553, + 587 + ], + "spans": [ + { + "bbox": [ + 317, + 544, + 553, + 587 + ], + "type": "text", + "content": "[24] Sheng Jin, Lumin Xu, Jin Xu, Can Wang, Wentao Liu, Chen Qian, Wanli Ouyang, and Ping Luo. Whole-body human pose estimation in the wild. In ECCV, pages 196-214. Springer, 2020. 3, 4" + } + ] + } + ], + "index": 27 + }, + { + "bbox": [ + 316, + 590, + 553, + 622 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 590, + 553, + 622 + ], + "spans": [ + { + "bbox": [ + 316, + 590, + 553, + 622 + ], + "type": "text", + "content": "[25] Trevor Johnston and Adam Schembri. Australian Sign Language (Auslan): An introduction to sign language linguistics. Cambridge University Press, 2007. 1" + } + ] + } + ], + "index": 28 + }, + { + "bbox": [ + 316, + 624, + 553, + 666 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 624, + 553, + 666 + ], + "spans": [ + { + "bbox": [ + 316, + 624, + 553, + 666 + ], + "type": "text", + "content": "[26] Hamid Reza Vaezi Joze and Oscar Koller. Ms-asl: A large-scale data set and benchmark for understanding american sign language. arXiv preprint arXiv:1812.01053, 2018. 2, 7" + } + ] + } + ], + "index": 29 + }, + { + "bbox": [ + 316, + 670, + 553, + 714 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 670, + 553, + 714 + ], + "spans": [ + { + "bbox": [ + 316, + 670, + 553, + 714 + ], + "type": "text", + "content": "[27] Alexander Kapitanov, Kvanchiani Karina, Alexander Nagaev, and Petrova Elizaveta. Slovo: Russian sign language dataset. In International Conference on Computer Vision Systems, pages 63-73. Springer, 2023. 2, 7, 8" + } + ] + } + ], + "index": 30 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "24328" + } + ] + } + ], + "index": 32 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 8 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 72, + 294, + 712 + ], + "type": "list", + "angle": 0, + "index": 15, + "blocks": [ + { + "bbox": [ + 56, + 72, + 294, + 105 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 72, + 294, + 105 + ], + "spans": [ + { + "bbox": [ + 56, + 72, + 294, + 105 + ], + "type": "text", + "content": "[28] Oscar Koller. Quantitative survey of the state of the art in sign language recognition. arXiv preprint arXiv:2008.09918, 2020. 2" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 106, + 294, + 149 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 106, + 294, + 149 + ], + "spans": [ + { + "bbox": [ + 56, + 106, + 294, + 149 + ], + "type": "text", + "content": "[29] Oscar Koller, Sepehr Zargaran, Hermann Ney, and Richard Bowden. Deep sign: Enabling robust statistical continuous sign language recognition via hybrid cnn-hmms. *IJCV*, 126: 1311–1325, 2018. 2" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 150, + 294, + 183 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 150, + 294, + 183 + ], + "spans": [ + { + "bbox": [ + 56, + 150, + 294, + 183 + ], + "type": "text", + "content": "[30] Taeryung Lee, Yeonguk Oh, and Kyoung Mu Lee. Human part-wise 3d motion context learning for sign language recognition. In ICCV, pages 20740-20750, 2023. 1, 6" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 56, + 184, + 294, + 227 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 184, + 294, + 227 + ], + "spans": [ + { + "bbox": [ + 56, + 184, + 294, + 227 + ], + "type": "text", + "content": "[31] Chao Li, Qiaoyong Zhong, Di Xie, and Shiliang Pu. Cooccurrence feature learning from skeleton data for action recognition and detection with hierarchical aggregation. arXiv preprint arXiv:1804.06055, 2018. 1, 3" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 56, + 228, + 294, + 282 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 228, + 294, + 282 + ], + "spans": [ + { + "bbox": [ + 56, + 228, + 294, + 282 + ], + "type": "text", + "content": "[32] Dongxu Li, Cristian Rodriguez, Xin Yu, and Hongdong Li. Word-level deep sign language recognition from video: A new large-scale dataset and methods comparison. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 1459-1469, 2020. 2, 7, 8" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 56, + 283, + 294, + 326 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 283, + 294, + 326 + ], + "spans": [ + { + "bbox": [ + 56, + 283, + 294, + 326 + ], + "type": "text", + "content": "[33] Dongxu Li, Xin Yu, Chenchen Xu, Lars Petersson, and Hongdong Li. Transferring cross-domain knowledge for video sign language recognition. In CVPR, pages 6205-6214, 2020. 6" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 56, + 327, + 294, + 370 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 327, + 294, + 370 + ], + "spans": [ + { + "bbox": [ + 56, + 327, + 294, + 370 + ], + "type": "text", + "content": "[34] Maosen Li, Siheng Chen, Xu Chen, Ya Zhang, Yanfeng Wang, and Qi Tian. Actional-structural graph convolutional networks for skeleton-based action recognition. In CVPR, pages 3595–3603, 2019. 3" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 56, + 371, + 294, + 425 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 371, + 294, + 425 + ], + "spans": [ + { + "bbox": [ + 56, + 371, + 294, + 425 + ], + "type": "text", + "content": "[35] Yanghao Li, Chao-Yuan Wu, Haoqi Fan, Karttikeya Mangalam, Bo Xiong, Jitendra Malik, and Christoph Feichtenhofer. Mviv2: Improved multiscale vision transformers for classification and detection. In CVPR, pages 4804-4814, 2022. 7, 8" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 56, + 426, + 294, + 470 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 426, + 294, + 470 + ], + "spans": [ + { + "bbox": [ + 56, + 426, + 294, + 470 + ], + "type": "text", + "content": "[36] Jiajia Liao, Yujun Liu, Yingchao Piao, Jinhe Su, Guorong Cai, and Yundong Wu. Gle-net: A global and local ensemble network for aerial object detection. International Journal of Computational Intelligence Systems, 15(1):2, 2022. 7" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 56, + 471, + 294, + 514 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 471, + 294, + 514 + ], + "spans": [ + { + "bbox": [ + 56, + 471, + 294, + 514 + ], + "type": "text", + "content": "[37] Jinfu Liu, Xinshun Wang, Can Wang, Yuan Gao, and Mengyuan Liu. Temporal decoupling graph convolutional network for skeleton-based gesture recognition. IEEE TMM, 26:811-823, 2023. 3" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 56, + 514, + 294, + 548 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 514, + 294, + 548 + ], + "spans": [ + { + "bbox": [ + 56, + 514, + 294, + 548 + ], + "type": "text", + "content": "[38] Ze Liu, Jia Ning, Yue Cao, Yixuan Wei, Zheng Zhang, Stephen Lin, and Han Hu. Video swim transformer. In CVPR, pages 3202-3211, 2022. 3, 7, 8" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 56, + 548, + 294, + 581 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 548, + 294, + 581 + ], + "spans": [ + { + "bbox": [ + 56, + 548, + 294, + 581 + ], + "type": "text", + "content": "[39] Anna Mindess. Reading between the signs: Intercultural communication for sign language interpreters. Nicholas Brealey, 2014. 1" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 56, + 582, + 294, + 624 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 582, + 294, + 624 + ], + "spans": [ + { + "bbox": [ + 56, + 582, + 294, + 624 + ], + "type": "text", + "content": "[40] Liliane Momeni, Hannah Bull, KR Prajwal, Samuel Albanie, Gül Varol, and Andrew Zisserman. Automatic dense annotation of large-vocabulary sign language videos. In ECCV, pages 671-690. Springer, 2022. 2" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 56, + 625, + 294, + 668 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 625, + 294, + 668 + ], + "spans": [ + { + "bbox": [ + 56, + 625, + 294, + 668 + ], + "type": "text", + "content": "[41] Evonne Ng, Shiry Ginosar, Trevor Darrell, and Hanbyul Joo. Body2hands: Learning to infer 3d hands from conversational gesture body dynamics. In CVPR, pages 11865-11874, 2021. 1, 3" + } + ] + } + ], + "index": 13 + }, + { + "bbox": [ + 56, + 669, + 294, + 712 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 669, + 294, + 712 + ], + "spans": [ + { + "bbox": [ + 56, + 669, + 294, + 712 + ], + "type": "text", + "content": "[42] Ladislav Rampášek, Michael Galkin, Vijay Prakash Dwivedi, Anh Tuan Luu, Guy Wolf, and Dominique Beaini. Recipe for a general, powerful, scalable graph transformer. NeurIPS, 35:14501-14515, 2022. 3" + } + ] + } + ], + "index": 14 + } + ], + "sub_type": "ref_text" + }, + { + "bbox": [ + 315, + 72, + 553, + 713 + ], + "type": "list", + "angle": 0, + "index": 33, + "blocks": [ + { + "bbox": [ + 315, + 72, + 553, + 106 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 72, + 553, + 106 + ], + "spans": [ + { + "bbox": [ + 315, + 72, + 553, + 106 + ], + "type": "text", + "content": "[43] Razieh Rastgoo, Kourosh Kiani, and Sergio Escalera. Sign language recognition: A deep survey. Expert Systems with Applications, 164:113794, 2021. 2" + } + ] + } + ], + "index": 16 + }, + { + "bbox": [ + 315, + 106, + 553, + 128 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 106, + 553, + 128 + ], + "spans": [ + { + "bbox": [ + 315, + 106, + 553, + 128 + ], + "type": "text", + "content": "[44] Wendy Sandler. Sign language and linguistic universals. Cambridge University, 2006. 1" + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 315, + 129, + 553, + 161 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 129, + 553, + 161 + ], + "spans": [ + { + "bbox": [ + 315, + 129, + 553, + 161 + ], + "type": "text", + "content": "[45] Noha Sarhan and Simone Frintrop. Unraveling a decade: a comprehensive survey on isolated sign language recognition. In ICCV, pages 3210-3219, 2023. 2" + } + ] + } + ], + "index": 18 + }, + { + "bbox": [ + 315, + 162, + 553, + 204 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 162, + 553, + 204 + ], + "spans": [ + { + "bbox": [ + 315, + 162, + 553, + 204 + ], + "type": "text", + "content": "[46] Lei Shi, Yifan Zhang, Jian Cheng, and Hanqing Lu. Two-stream adaptive graph convolutional networks for skeleton-based action recognition. In CVPR, pages 12026-12035, 2019. 3" + } + ] + } + ], + "index": 19 + }, + { + "bbox": [ + 315, + 205, + 553, + 248 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 205, + 553, + 248 + ], + "spans": [ + { + "bbox": [ + 315, + 205, + 553, + 248 + ], + "type": "text", + "content": "[47] Lei Shi, Yifan Zhang, Jian Cheng, and Hanqing Lu. Skeleton-based action recognition with multi-stream adaptive graph convolutional networks. IEEE TIP, 29:9532-9545, 2020. 2" + } + ] + } + ], + "index": 20 + }, + { + "bbox": [ + 315, + 250, + 553, + 293 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 250, + 553, + 293 + ], + "spans": [ + { + "bbox": [ + 315, + 250, + 553, + 293 + ], + "type": "text", + "content": "[48] Chenyang Si, Wentao Chen, Wei Wang, Liang Wang, and Tieniu Tan. An attention enhanced graph convolutional LSTM network for skeleton-based action recognition. In CVPR, pages 1227-1236, 2019. 3" + } + ] + } + ], + "index": 21 + }, + { + "bbox": [ + 315, + 294, + 553, + 326 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 294, + 553, + 326 + ], + "spans": [ + { + "bbox": [ + 315, + 294, + 553, + 326 + ], + "type": "text", + "content": "[49] Neil Song and Yu Xiang. Slgtformer: An attention-based approach to sign language recognition. arXiv preprint arXiv:2212.10746, 2022. 3, 4" + } + ] + } + ], + "index": 22 + }, + { + "bbox": [ + 315, + 327, + 553, + 381 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 327, + 553, + 381 + ], + "spans": [ + { + "bbox": [ + 315, + 327, + 553, + 381 + ], + "type": "text", + "content": "[50] Yi-Fan Song, Zhang Zhang, Caifeng Shan, and Liang Wang. Stronger, faster and more explainable: A graph convolutional baseline for skeleton-based action recognition. In proceedings of the 28th ACM international conference on multimedia, pages 1625-1633, 2020. 2" + } + ] + } + ], + "index": 23 + }, + { + "bbox": [ + 315, + 383, + 553, + 415 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 383, + 553, + 415 + ], + "spans": [ + { + "bbox": [ + 315, + 383, + 553, + 415 + ], + "type": "text", + "content": "[51] Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. Deep high-resolution representation learning for human pose estimation. In CVPR, pages 5693-5703, 2019. 3, 4" + } + ] + } + ], + "index": 24 + }, + { + "bbox": [ + 315, + 415, + 553, + 437 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 415, + 553, + 437 + ], + "spans": [ + { + "bbox": [ + 315, + 415, + 553, + 437 + ], + "type": "text", + "content": "[52] Clayton Valli. Linguistics of American Sign Language: An Introduction. Gallaudet University Press, 2000. 1" + } + ] + } + ], + "index": 25 + }, + { + "bbox": [ + 315, + 438, + 553, + 480 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 438, + 553, + 480 + ], + "spans": [ + { + "bbox": [ + 315, + 438, + 553, + 480 + ], + "type": "text", + "content": "[53] Gül Varol, Liliane Momeni, Samuel Albanie, Triantafyllos Afouras, and Andrew Zisserman. Scaling up sign spotting through sign language dictionaries. *IJCV*, 130(6):1416–1439, 2022. 2" + } + ] + } + ], + "index": 26 + }, + { + "bbox": [ + 315, + 481, + 553, + 524 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 481, + 553, + 524 + ], + "spans": [ + { + "bbox": [ + 315, + 481, + 553, + 524 + ], + "type": "text", + "content": "[54] Hongsong Wang and Liang Wang. Modeling temporal dynamics and spatial configurations of actions using two-stream recurrent neural networks. In CVPR, pages 499-508, 2017. 2" + } + ] + } + ], + "index": 27 + }, + { + "bbox": [ + 315, + 525, + 553, + 568 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 525, + 553, + 568 + ], + "spans": [ + { + "bbox": [ + 315, + 525, + 553, + 568 + ], + "type": "text", + "content": "[55] Hanjie Wang, Xiujuan Chai, and Xin Chen. A novel sign language recognition framework using hierarchical grassmann covariance matrix. IEEE TMM, 21(11):2806-2814, 2019. 2" + } + ] + } + ], + "index": 28 + }, + { + "bbox": [ + 315, + 569, + 553, + 613 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 569, + 553, + 613 + ], + "spans": [ + { + "bbox": [ + 315, + 569, + 553, + 613 + ], + "type": "text", + "content": "[56] Chengcheng Wei, Jian Zhao, Wengang Zhou, and Houqiang Li. Semantic boundary detection with reinforcement learning for continuous sign language recognition. IEEE TCSVT, 31(3):1138-1149, 2020." + } + ] + } + ], + "index": 29 + }, + { + "bbox": [ + 315, + 613, + 553, + 646 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 613, + 553, + 646 + ], + "spans": [ + { + "bbox": [ + 315, + 613, + 553, + 646 + ], + "type": "text", + "content": "[57] Pan Xie, Mengyi Zhao, and Xiaohui Hu. Pisltrc: Position-informed sign language transformer with content-aware convolution. IEEE TMM, 24:3908-3919, 2021. 2" + } + ] + } + ], + "index": 30 + }, + { + "bbox": [ + 315, + 647, + 553, + 680 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 647, + 553, + 680 + ], + "spans": [ + { + "bbox": [ + 315, + 647, + 553, + 680 + ], + "type": "text", + "content": "[58] Sijie Yan, Yuanjun Xiong, and Dahua Lin. Spatial temporal graph convolutional networks for skeleton-based action recognition. In AAAI, 2018. 1, 6, 7" + } + ] + } + ], + "index": 31 + }, + { + "bbox": [ + 315, + 681, + 553, + 713 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 681, + 553, + 713 + ], + "spans": [ + { + "bbox": [ + 315, + 681, + 553, + 713 + ], + "type": "text", + "content": "[59] Aoxiong Yin, Zhou Zhao, Weike Jin, Meng Zhang, Xingshan Zeng, and Xiaofei He. Mlslt: Towards multilingual sign language translation. In CVPR, pages 5109-5119, 2022. 1" + } + ] + } + ], + "index": 32 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 318, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 757 + ], + "type": "text", + "content": "24329" + } + ] + } + ], + "index": 34 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 9 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 72, + 294, + 240 + ], + "type": "list", + "angle": 0, + "index": 4, + "blocks": [ + { + "bbox": [ + 56, + 72, + 294, + 116 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 72, + 294, + 116 + ], + "spans": [ + { + "bbox": [ + 56, + 72, + 294, + 116 + ], + "type": "text", + "content": "[60] Weichao Zhao, Hezhen Hu, Wengang Zhou, Jiaxin Shi, and Houqiang Li. Best: Bert pre-training for sign language recognition with coupling tokenization. In AAAI, pages 3597-3605, 2023. 6, 7" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 118, + 294, + 161 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 118, + 294, + 161 + ], + "spans": [ + { + "bbox": [ + 56, + 118, + 294, + 161 + ], + "type": "text", + "content": "[61] Weichao Zhao, Hezhen Hu, Wengang Zhou, Yunyao Mao, Min Wang, and Houjiang Li. Masa: Motion-aware masked autoencoder with semantic alignment for sign language recognition. IEEE TCSVT, 2024. 6, 7" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 163, + 294, + 206 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 163, + 294, + 206 + ], + "spans": [ + { + "bbox": [ + 56, + 163, + 294, + 206 + ], + "type": "text", + "content": "[62] Weichao Zhao, Wengang Zhou, Hezhen Hu, Min Wang, and Houqiang Li. Self-supervised representation learning with spatial-temporal consistency for sign language recognition. arXiv preprint arXiv:2406.10501, 2024. 6, 7" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 56, + 208, + 294, + 240 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 208, + 294, + 240 + ], + "spans": [ + { + "bbox": [ + 56, + 208, + 294, + 240 + ], + "type": "text", + "content": "[63] Ronglai Zuo, Fangyun Wei, and Brian Mak. Natural language-assisted sign language recognition. In CVPR, pages 14890-14900, 2023. 6, 7" + } + ] + } + ], + "index": 3 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "text", + "content": "24330" + } + ] + } + ], + "index": 5 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 10 + } + ], + "_backend": "vlm", + "_version_name": "2.6.4" +} \ No newline at end of file diff --git a/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/4af7cfba-2431-4854-85f1-f3f911dcc811_content_list.json b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/4af7cfba-2431-4854-85f1-f3f911dcc811_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..9da3c96a10de250d68cb29d22b5ff79ca05d2e4c --- /dev/null +++ b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/4af7cfba-2431-4854-85f1-f3f911dcc811_content_list.json @@ -0,0 +1,1362 @@ +[ + { + "type": "text", + "text": "VTON 360: High-Fidelity Virtual Try-On from Any Viewing Direction", + "text_level": 1, + "bbox": [ + 142, + 130, + 852, + 152 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Zijian He1 Yuwei Ning1 Yipeng Qin2 Guangrun Wang1 Sibei Yang3 Liang Lin1,4,5 Guanbin Li1,4,5* \n1Sun Yat-sen University \n2Cardiff University \n3ShanghaiTech University \n4Guangdong Key Laboratory of Big Data Analysis and Processing \n5Peng Cheng Laboratory \nhezj39@mail2.sysu.edu.cn, yuwei.ning@hust.edu.cn, {qinyipengl991, wanggrun}@gmail.com, yangsb@shanghaiitech.edu.cn linliang@ieee.org, liguanbin@mail.sysu.edu.cn", + "bbox": [ + 99, + 159, + 898, + 250 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/3e4ed0ca9fe9c0466bbb1e4f4019077b1eea386c4b063ea02f3897e6315ec9c9.jpg", + "image_caption": [ + "Figure 1. Results of VTON 360. Our VTON 360 enables high-fidelity 3D Virtual Try-On (VTON) by seamlessly adapting E-commerce garments onto a clothed 3D human model, supporting full $360^{\\circ}$ view rendering. The highlighted bounding boxes (dashed line) demonstrate our method's ability to preserve intricate clothing details and patterns (e.g., collar accessories, horizontal line patterns, logos, texts, numbers) across diverse garment types." + ], + "image_footnote": [], + "bbox": [ + 91, + 252, + 509, + 474 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/18ca224d94844f9dde5541e3e34a54b63839bcf1d69569779a98633de60a2a1f.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 517, + 252, + 709, + 474 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/a5889a307af41522f81b308115e028b9c3f02dce5c03771e67bcdd87075b042d.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 714, + 252, + 906, + 474 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract", + "text_level": 1, + "bbox": [ + 246, + 542, + 326, + 558 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Virtual Try-On (VTON) is a transformative technology in e-commerce and fashion design, enabling realistic digital visualization of clothing on individuals. In this work, we propose VTON 360, a novel 3D VTON method that addresses the open challenge of achieving high-fidelity VTON that supports any-view rendering. Specifically, we leverage the equivalence between a 3D model and its rendered multiview 2D images, and reformulate 3D VTON as an extension of 2D VTON that ensures 3D consistent results across multiple views. To achieve this, we extend 2D VTON models to include multi-view garments and clothing-agnostic human body images as input, and propose several novel techniques to enhance them, including: i) a pseudo-3D pose representation using normal maps derived from the SMPL-X 3D human model, ii) a multi-view spatial attention mechanism that models the correlations between features from different viewing angles, and iii) a multi-view CLIP embedding that enhances the garment CLIP features used in 2D VTON with camera information. Extensive experiments on large-scale real datasets and clothing images from e-commerce", + "bbox": [ + 88, + 569, + 485, + 872 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "platforms demonstrate the effectiveness of our approach. \nProject page: https://scnuhealthy.github.io/ \nVTON360.", + "bbox": [ + 511, + 544, + 906, + 589 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1. Introduction", + "text_level": 1, + "bbox": [ + 513, + 601, + 643, + 616 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Virtual Try-On (VTON) enables realistic digital visualization of clothing on individuals and has emerged as a transformative technology in e-commerce and fashion design. While significant research efforts have been made on 2D VTON solutions [8, 12, 19, 33, 39], these approaches are inherently limited in their representation of view-related features. To overcome this limitation and enable high-fidelity any-view rendering, 3D VTON methods were introduced.", + "bbox": [ + 509, + 627, + 906, + 748 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "3D VTON requires accurate garment transfer onto a 3D human body while ensuring realistic garment fitting, texture preservation, and 3D consistency. The two primary aims of 3D VTON are i) achieving high-fidelity and ii) supporting any-view rendering. Leveraging the inherent capability of 3D models for any-view rendering, early 3D VTON methods [13, 15, 28] make clothing simulation on synthetic human bodies. Specifically, these methods utilized 3D scanners to capture clothing meshes, followed by the development of specialized dressing algorithms. Although effec", + "bbox": [ + 509, + 750, + 908, + 901 + ], + "page_idx": 0 + }, + { + "type": "header", + "text": "CVF", + "bbox": [ + 106, + 2, + 181, + 42 + ], + "page_idx": 0 + }, + { + "type": "header", + "text": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore.", + "bbox": [ + 236, + 0, + 807, + 46 + ], + "page_idx": 0 + }, + { + "type": "page_footnote", + "text": "* Corresponding author is Guanbin Li.", + "bbox": [ + 109, + 886, + 310, + 898 + ], + "page_idx": 0 + }, + { + "type": "page_number", + "text": "26388", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "tive, these methods rely on costly 3D scanning equipment and the physical presence of the human body/clothing (i.e., not fully virtual), restricting their practicality in real-world applications. As a byproduct, most early methods focused on developing geometrically correct dressing algorithms using standard templates of human body and clothing models. Addressing this limitation, researchers extended 3D VTON by introducing algorithms that reconstruct 3D clothing models from input images, enabling the use of image-based clothing inputs [3, 32, 41, 42]. However, since input clothing images (usually frontal) are inherently 2D and lack multi-view information, this approach struggles to reconstruct high-fidelity clothing models that can be rendered well from all viewing directions.", + "bbox": [ + 89, + 90, + 480, + 303 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "To complement this missing information, DreamVTON [50] introduces a novel approach that leverages Text-to-Image (T2I) diffusion models to reconstruct both the human body and clothing from input images. Its key insight is that T2I models learned view-agnostic \"concepts\" of both bodies and garments during their training, and that the corresponding concepts for the input body and clothing images can be obtained using LoRA [21]. By utilizing Score Distillation Sampling (SDS) [37], DreamVTON can generate visual-pleasing 3D VTON results by ensuring consistency between renderings from arbitrary viewpoints and the concepts. Nonetheless, DreamVTON's high flexibility comes at the cost of low fidelity. This limitation stems from the fact that the concepts learned by T2I models are semantic in nature, thus lacking 3D geometric consistency and pixel-level accuracy with respect to the input body and clothing images. Recently, a concurrent work, namely GaussianVTON [6], partially addressed this limitation by formulating 3D VTON as a 3D scene editing task, where a given 3D human model is edited using multi-view images generated by 2D VTON methods. While it significantly enhances the fidelity of the human body, the fidelity and 3D consistency of clothing remain problematic, as there are no 2D VTON methods that can generate multi-view images with 3D consistency. Therefore, to the best of our knowledge, achieving high-fidelity 3D VTON that supports any-view rendering remains an open challenge.", + "bbox": [ + 91, + 306, + 482, + 715 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this work, we address the above-mentioned challenge via proposing VTON 360, a novel 3D VTON method that achieves high-fidelity VTON from arbitrary viewing directions. Similar to GaussianVTON [6], our method edits a given 3D human model by inpainting the rendered images using a latent diffusion model. However, we set ourselves apart through our novel garment fidelity preservation strategy that can generate high-fidelity on-body garments in all viewing directions. Specifically, we first extend both the garment and clothing-agnostic human body inputs to typical 2D VTON models to leverage multi-view information, including paired front and back view garment images as well", + "bbox": [ + 89, + 719, + 482, + 901 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "as a set of multi-view clothing-agnostic human body images sampled from random azimuth angles. Then, we propose several novel enhancements to bridge the gap between typical 2D VTON methods and our multi-view 3D consistency requirements: i) We propose a pseudo-3D pose representation using normal maps derived from the SMPL-X 3D human model, which captures fine-grained surface orientation details and provides more consistent geometry across views compared to the 2D pose representations (semantic segmentation maps) used in 2D VTON models. ii) We design a Multi-view Spatial Attention mechanism that models the correlations between features from different viewing angles, featuring a novel \"correlation\" matrix modeling the relationships among different input views. iii) We propose a multi-view CLIP embedding that enhances the garment CLIP embedding used in 2D VTON methods with camera information, thereby facilitating network learning of features relevant to a particular view. Together, these innovations enable our 2D VTON model to generate high-quality, multi-view and 3D-consistent virtual try-on results. Extensive experiments on Thuman2.0 [55] and MVHumanNet [51] datasets demonstrate that our method achieves high fidelity 3D VTON which supports any-view rendering. In addition, we show the effectiveness and generalizability of our methodology by testing it using garments from e-commerce platforms. Our conclusions include:", + "bbox": [ + 511, + 90, + 903, + 482 + ], + "page_idx": 1 + }, + { + "type": "list", + "sub_type": "text", + "list_items": [ + "- We propose a novel 3D Virtual Try-On (VTON) method, namely VTON 360, which achieves high-fidelity VTON from arbitrary viewing directions.", + "- Leveraging the equivalence between a 3D model and its rendered multi-view 2D images, we reformulate 3D VTON as an extension of 2D VTON that ensures consistent results across multiple views. Specifically, we introduce several novel techniques, including: (i) pseudo-3D pose representation; (ii) multi-view spatial attention; and (iii) multi-view CLIP embedding. These innovations enhance traditional 2D VTON models to generate multiview and 3D-consistent results.", + "- Extensive experimental results on two large real datasets as well as real clothing images from e-commerce platforms demonstrate the effectiveness of our approach." + ], + "bbox": [ + 513, + 483, + 903, + 710 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2. Related Work", + "text_level": 1, + "bbox": [ + 513, + 723, + 653, + 739 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2D Virtual Try-On. 2D Virtual Try-On (VTON) aims to transfer a desired garment to the corresponding region of a target human image while preserving the human pose and identity. Early methods [2, 8, 10, 11, 16, 18, 29, 31, 38, 54, 56] use Generative Adversarial Networks (GANs) to deform the garments to match the target body shape, which a critical step for achieving realistic VTON. However, accurately adapting to diverse real-world conditions remains a significant challenge. Addressing this issue, recent methods [12, 19, 33, 60] reframe 2D VTON as a conditioned in-", + "bbox": [ + 511, + 750, + 903, + 900 + ], + "page_idx": 1 + }, + { + "type": "page_number", + "text": "26389", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "painting task, leveraging the strong priors provided by diffusion models [20, 43, 45] to achieve promising results. This strategy is further improved by [9, 26, 53], which introduce a ReferenceNet to extract hierarchical garment features and apply attention mechanisms to condition the Main UNet.", + "bbox": [ + 89, + 90, + 480, + 167 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3D Virutal Try-On. For 3D Virtual Try-On (VTON), traditional methods [4, 13, 15, 28, 36] rely on 3D scanning or cloth simulation to generate highly precise body and garment geometry. These methods were then extended by learning-based methods [3, 32] that employ differentiable rendering to dress the SMPL [30] model with a desired garment mesh. Despite their effectiveness, such methods rely on costly 3D scanning and the physical presence of human body/clothing, limiting their application in the real world. Addressing this limitation, M3D-VTON [59] proposes a depth-based 3D VTON framework to reconstruct 3D clothed human models from 2D human and garment images, but the results often suffer from explicit warping artifacts. To improve 3D VTON results, recent methods [23, 24, 50, 62] resort to text-to-image (T2I) diffusion models and employ the Score Distillation Sampling (SDS) loss [40] to ensure consistency among different viewing directions. Specifically, TeCH [24] adapts the generative priors of T2I diffusion model to the specific person and clothes by training descriptive text prompts with DreamBooth [40]. DreamWaltz [23] leverages Pose ControlNet [57] to attain clothed human body models. DreamVTON [50] introduces a multi-concept LoRA [21] to personalize the T2I diffusion model, and uses a template-based optimization mechanism that combines with SDS loss to better preserve patterns on the garment. Although effective, these methods often produce results lacking in fidelity, as the concepts learned by T2I models are semantic rather than at the pixel level. Concurrent to our work, GaussianVTON [6] proposes an alternative approach by combining Gaussian Splatting [25] with pre-trained 2D VTON models and formulate it as an editing task. However, since there are no 2D VTON methods that can generate multi-view images with 3D consistency, the fidelity and 3D consistency of the clothing generated remain problematic.", + "bbox": [ + 91, + 178, + 483, + 708 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Radiance Field-based 3D Human or Scene Editing. Recently, radiance field-based editing has attracted significant interest due to its efficient differentiable rendering capabilities, sparking substantial advancements in text-driven 3D editing. For example, InstructN2N [17] employ an image-based diffusion model InstructP2P [5] to modify the rendered image by the user's text description with a variant of the score distillation sampling (SDS) [37] loss. GaussianEditor [7] applies Gaussian Splatting [25] as 3D representation instead of NeRF, adopting Gaussian semantic tracking to track target Gaussian values, significantly improving editing speed and controllability. To enable accu", + "bbox": [ + 89, + 719, + 482, + 900 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "rate location and appearance control, subsequent works [47, 61] specify the editing region using the attention score or with a segmentation model. TIP-Editor [63] proposes a content personalization step dedicated to the reference image based on LoRA, achieving the editing following a hybrid text-image prompt. GaussCtrl [48] leverage depth conditions and attention-based latent code alignment to achieve 3D-aware multi-view consistent editing instead of iteratively editing single views using SDS loss. However, these works primarily focus on global appearance modifications within a text-driven pipeline, while our approach emphasizes preserving fine textural details from different viewing directions throughout the editing process.", + "bbox": [ + 511, + 90, + 906, + 287 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3. Preliminary", + "text_level": 1, + "bbox": [ + 511, + 300, + 638, + 316 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Latent Diffusion Model. Latent Diffusion Model [39] is a variant of diffusion models that performs denoising within the latent space of a Variational Autoencoder (VAE) [27]. During training, given a fixed encoder $\\mathcal{E}$ , an input image $x$ is transformed into its latent representation $z_0 = \\mathcal{E}(x)$ . A conditional diffusion model $\\hat{\\epsilon}_{\\theta}$ , typically implemented with a UNet architecture, is then trained using a weighted denoising score matching objective:", + "bbox": [ + 511, + 325, + 906, + 446 + ], + "page_idx": 2 + }, + { + "type": "equation", + "text": "\n$$\n\\mathcal {L} _ {L D M} = \\mathbb {E} _ {\\mathbf {z}, \\mathbf {c}, \\boldsymbol {\\epsilon}, t} [ \\boldsymbol {\\epsilon} - \\| \\hat {\\epsilon} _ {\\theta} (\\mathbf {z} _ {\\mathbf {t}}; \\mathbf {c}, t) \\| _ {2} ^ {2} ] \\tag {1}\n$$\n", + "text_format": "latex", + "bbox": [ + 583, + 455, + 906, + 474 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\mathbf{z}_t\\coloneqq \\alpha_t\\mathbf{x} + \\sigma_t\\boldsymbol {\\epsilon}$ denotes the forward diffusion process at timestep $t$ . $\\alpha_{t},\\sigma_{t}$ are time-dependent functions defined by the diffusion model formulation; c denotes the conditional input and $\\epsilon \\sim \\mathcal{N}(\\mathbf{0},\\mathbf{1})$ is Gaussian noise. During inference, data samples are generated by initiating from Gaussian noise $\\mathbf{z}_T\\sim \\mathcal{N}(\\mathbf{0},\\mathbf{1})$ and iteratively refining it using a DDIM [44] sampler.", + "bbox": [ + 511, + 484, + 908, + 590 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "4. Method", + "text_level": 1, + "bbox": [ + 511, + 603, + 604, + 618 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Our method leverages the equivalence between a 3D model and its rendered multi-view 2D images to achieve high-fidelity, any-view 3D VTON. Specifically, as Fig. 2 shows, given an input 3D human model and a garment image, our method 1) renders the 3D model into multi-view 2D images and 2) formulates 3D VTON as a consistent, unified 2D VTON process across these rendered views; 3) By reconstructing the edited images into a 3D model using existing 3D reconstruction methods, we ensure visual coherence and precise garment alignment from any viewing angle. Among them, the second step is crucial as existing 2D VTON methods [9, 26, 53] lack 3D knowledge, preventing them from generating multi-view images with 3D consistency.", + "bbox": [ + 511, + 628, + 908, + 824 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "To address this challenge, we propose several novel techniques (Sec. 4.2) that equip a typical 2D VTON network (Sec. 4.1), which is built on a latent diffusion model [39], with the capability to generate 3D-consistent results. We use Gaussian Splatting [25] as our 3D representation.", + "bbox": [ + 511, + 825, + 906, + 902 + ], + "page_idx": 2 + }, + { + "type": "page_number", + "text": "26390", + "bbox": [ + 478, + 944, + 519, + 957 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/50456d17e5ccbb76f3e583fe2b0ebac3fa363e2d03f09e2a13d08f82ca40471e.jpg", + "image_caption": [ + "Figure 2. Overview of VTON 360. Given an input 3D human model $\\mathbf{G}_{\\mathrm{src}}$ and a pair of garment images $(g_{f}, g_{b})$ , our method 1) renders $\\mathbf{G}_{\\mathrm{src}}$ into multi-view 2D images (left) and 2) edits the rendered multi-view 2D images (middle); 3) reconstructs the edited images into a 3D model $\\mathbf{G}_{\\mathrm{VTON}}$ (right). In the crucial step 2), we propose three novel techniques to equip a typical 2D VTON network with the capability to generate 3D-consistent results: 1) Pseudo-3D Pose Input, 2) Multi-view Spatial Attention, and 3) Multi-view CLIP Embedding." + ], + "image_footnote": [], + "bbox": [ + 89, + 88, + 903, + 431 + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/c3dd68bc18957e8ae4d2aa15c28ee2931189de321d4ce4c838cc8d0a92c0acbb.jpg", + "image_caption": [ + "Figure 3. DensePose (2D) vs. SMPL-X normal map (pseudo-3D) representations. DensePose applies uniform labels per body part, lacking 3D consistency across views and causing artifacts and temporal inconsistencies (highlighted with red boxes). In contrast, SMPL-X normal maps capture fine surface details, ensuring geometric coherence and stable, realistic shading across views." + ], + "image_footnote": [], + "bbox": [ + 91, + 503, + 480, + 679 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "4.1. Recap of 2D VTON Framework", + "text_level": 1, + "bbox": [ + 89, + 784, + 372, + 800 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Following previous works [12, 26, 53], we formulate 2D VTON as an exemplar-based inpainting problem, aiming to fill an input clothing-agnostic image $\\mathbf{A}$ with a given garment image $g$ , where $\\mathbf{A}$ is obtained following the method used in [53]. As illustrated in Fig. 2 (middle), the network architecture is based on the latent diffusion model [39] with", + "bbox": [ + 89, + 809, + 483, + 900 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "an encoder $\\mathcal{E}$ and comprises two components:", + "bbox": [ + 513, + 506, + 816, + 520 + ], + "page_idx": 3 + }, + { + "type": "list", + "sub_type": "text", + "list_items": [ + "- A GarmentNet [9, 53] that extracts features from $\\mathcal{E}(g)$ .", + "- A Main UNet that inpaints $\\mathbf{A}$ according to i) detailed garment features extracted by the GarmentNet; ii) the 2D pose of $\\mathbf{A}$ represented by semantic labels using Dense-Pose [14]; iii) CLIP embeddings of input garment $g$ . Among them, i) and ii) together with noise are input to the self-attention layers, while iii) is input to the cross-attention layers of the Main UNet." + ], + "bbox": [ + 511, + 520, + 903, + 640 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Both the GarmentNet and the Main UNet share the same network architecture.", + "bbox": [ + 511, + 641, + 903, + 670 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "4.2. Multi-view 2D VTON with 3D Consistency", + "text_level": 1, + "bbox": [ + 511, + 681, + 877, + 698 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "To enable the aforementioned 2D VTON model to generate multi-view and 3D-consistent results, we propose the following novel enhancements to its design:", + "bbox": [ + 511, + 704, + 903, + 750 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Multi-view Inputs. We extend both inputs to the model:", + "bbox": [ + 513, + 750, + 887, + 763 + ], + "page_idx": 3 + }, + { + "type": "list", + "sub_type": "text", + "list_items": [ + "- Multi-view Garment Inputs: We extend the input garment representation from a single image $g$ to paired front and back view images $g_{f}$ , $g_{b}$ , providing comprehensive garment information across all viewing angles. Accordingly, we use the encoder $\\mathcal{E}$ to encode $g_{f}, g_{b}$ into their latent representations $\\mathcal{E}(g_{f}), \\mathcal{E}(g_{b})$ and feed them into GarmentNet to obtain their multi-layer features $F_{f}^{l}$ and $F_{b}^{l}$ at layer $l$ , respectively.", + "- Multi-view Clothing-agnostic Image Inputs: Based on" + ], + "bbox": [ + 513, + 765, + 905, + 901 + ], + "page_idx": 3 + }, + { + "type": "page_number", + "text": "26391", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "the equivalence between a 3D human model and its rendered multi-view 2D images, we extend the input human body representation from a single, clothing-agnostic image, $\\mathbf{A}$ , to a set of $m$ multi-view images, denoted as $\\mathbf{A}_1, \\mathbf{A}_2, \\dots, \\mathbf{A}_m$ . These images are sampled from random azimuth angles, allowing the 2D VTON model to access comprehensive, multi-view information from the input 3D human model.", + "bbox": [ + 102, + 90, + 482, + 210 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Pseudo-3D Pose Input. As shown in Fig. 3, the 2D Dense-Pose representations [14] commonly used in state-of-the-art 2D VTON methods [9, 26] assign a uniform semantic label to all pixels within each body part (e.g., thigh), inherently lack 3D geometric consistency across multiple views, and often introduce artifacts and temporal inconsistencies. To address these limitations, we propose a novel pseudo-3D pose representation: the normal maps $\\mathbf{N}$ derived from the SMPL-X [35] model of the input body. These normal maps capture fine-grained surface orientation details, providing a more consistent representation across views by preserving geometric structure in the 3D space. Furthermore, they enable smoother, temporally stable transitions and realistic shading effects across multi-view scenarios. In practice, we employ a lightweight PoseEncoder $\\mathcal{E}'$ [22] and feed $\\mathcal{E}'(\\mathbf{N})$ into the Main UNet. We obtain the SMPL-X model from the multi-view images of the input body using EasyMoCap [1].", + "bbox": [ + 89, + 212, + 482, + 469 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Accordingly, we concatenate three components as the enhanced input to the Main UNet: i) a noise latent $z_{t}$ ; ii) the encoded pseudo-3D poses $\\mathcal{E}'(\\mathbf{N_1}),\\mathcal{E}'(\\mathbf{N_2}),\\dots,\\mathcal{E}'(\\mathbf{N_m})$ ; and iii) the encoded multi-view clothing-agnostic images $\\mathcal{E}(\\mathbf{A_1}),\\mathcal{E}(\\mathbf{A_2}),\\dots,\\mathcal{E}(\\mathbf{A_m})$ . Let $F_{1}^{l},F_{2}^{l},\\dots,F_{m}^{l}$ be the feature representations at layer $l$ of the Main UNet, and recall the garment features $F_{f}^{l}$ and $F_{b}^{l}$ defined above, we enhance the self-attention layers of the Main UNet as:", + "bbox": [ + 89, + 470, + 482, + 589 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Multi-view Spatial Attention. To cope with the aforementioned multi-view input features and ensure their consistency, we draw insights from the temporal attention layer commonly used in video generation and editing [49, 58] and extend it to our multi-view spatial attention layer, denoted as MVAttention. The key distinction of our MVAttention is that its input multi-view features $F_{1}^{l}, F_{2}^{l}, \\ldots, F_{m}^{l}$ are from images captured from non-uniform spatial intervals, with the viewing angles varying randomly. Consequently, features from similar views exhibit a higher correlation, while those from distinct views are largely independent. To model this relationship, we construct a \"correlation\" matrix $C$ based on the angular disparity obtained from camera rotation matrices of the input multi-view images, and define our MVAttention as follows:", + "bbox": [ + 89, + 590, + 482, + 816 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\n\\begin{array}{l} \\mathbf {F} ^ {\\mathbf {l}} = \\left[ F _ {1} ^ {l} \\oplus F _ {2} ^ {l} \\dots \\oplus F _ {m} ^ {l} \\right], \\hat {\\mathbf {F}} ^ {\\mathbf {l}} = \\left[ \\mathbf {F} ^ {\\mathbf {l}} \\oplus F _ {f} ^ {l} \\oplus F _ {b} ^ {l} \\right] \\\\ Q = W ^ {Q} \\mathbf {F} ^ {1}, K = W ^ {K} \\hat {\\mathbf {F}} ^ {1}, V = W ^ {V} \\hat {\\mathbf {F}} ^ {1} \\tag {2} \\\\ A _ {i} = \\operatorname {s o f t m a x} \\left(\\frac {Q _ {i} \\times \\left(C _ {i} \\cdot K ^ {T}\\right)}{\\sqrt {d}}\\right), \\mathbf {H} _ {\\mathbf {i}} ^ {\\mathbf {l}} = A _ {i} \\times V _ {i} \\\\ \\end{array}\n$$\n", + "text_format": "latex", + "bbox": [ + 112, + 825, + 480, + 905 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $i\\in \\{1,2,\\dots,m\\}$ denotes $i$ -th view; the Query $Q$ comes directly from $\\mathbf{F}^1$ and the concatenation of $[\\mathbf{F}^1,F_f^l,F_b^l ]$ serves as the key $K$ and the value $V$ . $\\oplus$ indicates matrix concatenation along the token axis; $d$ denotes the dimension; $W^{Q},W^{K},W^{V}$ represent the linear transformation matrices; we omitted the $l$ of the attention matrices and parameters for simplicity; $C\\in \\mathbb{R}^{m\\times m}$ $C_i$ represents $i$ -th row in $C$ , and its \"correlation\" value between $i$ -th and $j$ -th features is $C_{ij}$ ..", + "bbox": [ + 511, + 90, + 903, + 227 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nC _ {i j} = \\left(\\left(\\operatorname {t r a c e} \\left(R _ {i} ^ {T} R _ {j}\\right) - 1\\right) / 2 + 1\\right) / 2 \\tag {3}\n$$\n", + "text_format": "latex", + "bbox": [ + 583, + 239, + 903, + 258 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $R_{i}$ and $R_{j}$ are the extrinsic rotation matrices of the corresponding camera views, $(\\mathrm{trace}(R_i^T R_j) - 1) / 2$ is the cosine value of the angle between these camera views.", + "bbox": [ + 511, + 268, + 903, + 314 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Multi-view CLIP Embedding. Camera viewpoints can serve as an effective condition signal to enhance 3D consistency in video content generation [52]. Building on this insight, we incorporate camera condition within our try-on network by encoding camera parameters as an additional token, enabling the generation of more consistent multi-view images. Specifically, we define a world coordinate system in which the camera faces the subject directly. For each input image (view) $\\mathbf{A}_{\\mathrm{i}}$ , $1 \\leq i \\leq m$ , we extract the rotation matrix from the camera's corresponding extrinsic matrix. This rotation matrix is then reshaped into a 9-dimensional tensor $\\mathbf{r}_{\\mathrm{i}}$ , which undergoes positional encoding to effectively integrate the camera parameters into the feature representation $F_{i}^{c}$ .", + "bbox": [ + 511, + 315, + 905, + 527 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\n\\begin{array}{l} F _ {i} ^ {c} = \\left(\\sin \\left(2 ^ {0} \\pi \\mathbf {r} _ {\\mathbf {i}}\\right), \\cos \\left(2 ^ {0} \\pi \\mathbf {r} _ {\\mathbf {i}}\\right), \\dots , \\right. \\tag {4} \\\\ \\sin (2 ^ {L - 1} \\pi \\mathbf {r _ {i}}), \\cos (2 ^ {L - 1} \\pi \\mathbf {r _ {i}})) \\\\ \\end{array}\n$$\n", + "text_format": "latex", + "bbox": [ + 594, + 539, + 903, + 578 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $L$ is the length of positional embedding. We then project $F_{i}^{c}$ to match the dimensionality of the garment CLIP image embedding $F^{g}$ via an MLP and concatenate them along the token axis to form $Y_{i}$ . This combined representation, $Y_{i}$ , is subsequently used in the key $K_{\\mathrm{x}}$ and value $V_{\\mathrm{x}}$ of the cross-attention layers of the Main UNet:", + "bbox": [ + 511, + 590, + 905, + 679 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nY _ {i} = F ^ {g} \\oplus \\operatorname {M L P} \\left(F _ {i} ^ {c}\\right)\n$$\n", + "text_format": "latex", + "bbox": [ + 635, + 693, + 777, + 709 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nQ _ {\\mathrm {x}} = W _ {\\mathrm {x}} ^ {Q} \\mathbf {H} _ {\\mathrm {i}} ^ {1}, K _ {\\mathrm {x}} = W _ {\\mathrm {x}} ^ {K} Y _ {i}, V _ {\\mathrm {x}} = W _ {\\mathrm {x}} ^ {V} Y _ {i} \\tag {5}\n$$\n", + "text_format": "latex", + "bbox": [ + 560, + 712, + 903, + 736 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nF _ {i} ^ {(l + 1)} = \\mathrm {s o f t m a x} (\\frac {Q _ {\\mathrm {x}} K _ {\\mathrm {x}} ^ {T}}{\\sqrt {d _ {\\mathrm {x}}}}) V _ {\\mathrm {x}},\n$$\n", + "text_format": "latex", + "bbox": [ + 586, + 734, + 794, + 768 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $\\mathbf{H}^1$ is the output of the MVAttention of the $l$ -th layer; we omitted the $l$ of the cross attention matrices and parameters for simplicity.", + "bbox": [ + 511, + 779, + 903, + 825 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Training. Our enhanced multi-view 2D VTON network can be trained by minimizing the following latent diffusion model loss function:", + "bbox": [ + 511, + 825, + 903, + 869 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\n\\mathcal {L} _ {\\mathrm {l d m}} = \\mathbb {E} _ {z _ {t}, \\eta , \\psi , \\epsilon , t} \\left[ \\| \\epsilon - \\epsilon_ {\\theta} (z _ {t}, t, \\eta , \\psi , \\zeta)) \\| _ {2} ^ {2} \\right], \\tag {6}\n$$\n", + "text_format": "latex", + "bbox": [ + 553, + 883, + 903, + 902 + ], + "page_idx": 4 + }, + { + "type": "page_number", + "text": "26392", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/66efb9bdfdf378e1471d23c74712a4482299aa59ce3a5ebfeba941baa74bac58.jpg", + "image_caption": [ + "Figure 4. Illustration of the proposed Multi-view Spatial Attention. Query (Q): multi-view features $\\mathbf{F}^1$ ; Key (K) and Value (V): concatenation of $\\mathbf{F}^1$ and garment features $F_{f}^{l}$ and $F_{b}^{l}$ . The attention score between viewpoints $i$ and $j$ is modulated by a weight $C_{ij}$ , determined by the cosine of the angle between them." + ], + "image_footnote": [], + "bbox": [ + 93, + 88, + 486, + 294 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $\\eta = [\\mathcal{E}(g_f); \\mathcal{E}(g_b); \\mathcal{E}(\\mathbf{N_i})_{i=1}^m]$ represents the input latent garment images and latent normal maps; $\\zeta = [\\mathcal{E}'(\\mathbf{A_i})_{i=1}^m]$ denotes the input latent clothing-agnostic images; $\\psi = \\mathbf{Y}$ is the proposed multi-view CLIP embedding.", + "bbox": [ + 89, + 380, + 483, + 443 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "5. Experiments", + "text_level": 1, + "bbox": [ + 89, + 455, + 225, + 474 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "5.1. Experimental Setup", + "text_level": 1, + "bbox": [ + 89, + 481, + 282, + 500 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Datasets. We conduct experiments on two public datasets: Thuman2.0 [55] and MVHumanNet [51]. Thuman2.0 comprises 526 reconstructed clothed human scans, from which we render multi-view input images. Of these samples, 426 are used for training, while the remaining 100 are set aside for testing. To further evaluate the effectiveness and robustness of our method, we also perform experiments on MVHumanNet, a large-scale dataset of multi-view human images that encompasses a diverse range of subjects, daily outfits and motion sequences. The images in MVHumanNet are captured using a multi-view system with either 48 or 24 cameras. We use 4,990 subjects from this dataset, allocating 4,790 to training and 200 for tests. For each subject, we randomly select two frames of multi-view images from its entire motion sequence. While MVHumanNet provides multi-view images directly for editing and reconstruction, we render uniformly distributed views around each human subject in Thuman2.0 to ensure consistent input.", + "bbox": [ + 88, + 503, + 482, + 775 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Baselines. We primarily compare our method with three existing methods: DreamWaltz [23], GaussCtrl [48], and TIP-Editor [63]. DreamWaltz is a method designed for directly generating 3D human bodies based on textual descriptions, while GaussCtrl and TIP-Editor are two radiance-based editing methods. GaussCtrl is based on Stable Diffusion, using a description-like prompt to edit the scene. TIP-Editor accepts both text and image prompts. We configure it by", + "bbox": [ + 89, + 780, + 482, + 902 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "specifying the human body as the editing region and the desired garment as the image prompt. We use ChatGPT to generate the text prompts corresponding to the clothing images.", + "bbox": [ + 511, + 90, + 903, + 151 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Evaluation Metrics. For quantitative evaluation, we assess garment-to-person alignment between the edited person and the reference image. Following [63], we calculate the average DINO similarity [34] between the reference image and the rendered multi-view images of the edited 3D scene. Additionally, to evaluate multi-view consistency, we compute the CLIP Directional Consistency Score as outlined in [17]. Given the large scale of experiments (repeated 3DGS reconstruction), we selected a subset of examples from the dataset for metric evaluation. Specifically, from the test sets of Thuman and MVHumanNet, we randomly sampled 10 human scans each, performing virtual try-on with 6 randomly chosen garments per human scan.", + "bbox": [ + 511, + 155, + 906, + 351 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We further conducted a user study involving 50 participants who rated the results of our method and three baseline methods based on two criteria: overall \"Quality\" and \"Alignment\" with the reference image. Each evaluation consisted of two questions: (1) Which method produces the highest quality of the edited 3D human? and (2) Which method achieves the most consistent alignment with the target clothing? Participants viewed the VTON results as rotating randomized video sequences.", + "bbox": [ + 511, + 351, + 908, + 488 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Implementation Details. During pre-processing, we crop the multi-view images to the bounding box around the person and resize them to a resolution of $768 \\times 576$ . The front view and the back view of garment images are obtained from the corresponding clothed human images. After editing, we pad the images back to their original size. The data processing pipeline is the same for both Thuman2.0 and MVHumanNet datasets.", + "bbox": [ + 511, + 491, + 908, + 611 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "The Main UNet and the GarmentNet are initialized by the Stable Diffusion V1.5 [39]. The training process is divided into two stages. In the first stage, each view is trained independently, during which we establish the feature extraction capabilities of both the PoseEncoder and GarmentNet, as well as the generative capability of the Main UNet. The second stage involves multi-view training, where we randomly select $M$ views for each human subject. This stage is focused on training the proposed MVAttention module to enhance the network's ability to maintain consistency across views. Due to memory constraints, we set $M = 8$ for the training phase. During the testing phase, we uniformly sampled 32 views from a 360-degree rotation around the subject. The editing of these 32 views is conducted in two batches, with each batch processing $M = 16$ views.", + "bbox": [ + 511, + 612, + 908, + 839 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "5.2. Comparisons with State-of-the-Art Methods", + "text_level": 1, + "bbox": [ + 511, + 848, + 888, + 864 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Qualitative Evaluation. Fig. 5 shows visual comparisons between our method and the baselines. DreamWaltz [23]", + "bbox": [ + 511, + 869, + 906, + 902 + ], + "page_idx": 5 + }, + { + "type": "page_number", + "text": "26393", + "bbox": [ + 478, + 944, + 519, + 957 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/16b7e1376295cee170255b3f0d8c6c37631ea1766a2dacc28c5a68023f4e5152.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
MethodThuman2.0 [55]MVHumanNet [51]
CLIPcons↑DINOsim↑VotequalityVotealignCLIPcons↑DINOsim↑VotequalityVotealign
DreamWaltz [23]0.8870.5560.46%1.54%0.9350.4950.46%0.46%
TIP-Editor [63]0.9390.5690.92%0.62%0.9480.5122.15%1.38%
GaussCtrl [48]0.9310.5771.08%1.38%0.9380.5211.69%1.23%
Ours0.9230.63397.54%96.46%0.9330.62395.69%96.92%
", + "bbox": [ + 96, + 88, + 901, + 180 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Table 1. Quantitative comparisons. ${\\mathrm{{CLIP}}}_{\\text{cons }}$ denotes the CLIP Direction Consistency Score. ${\\mathrm{{DINO}}}_{\\text{sim }}$ is the DINO similarity.", + "bbox": [ + 114, + 185, + 879, + 200 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/6b71985790ddfa2ccbcfe297ed7bf36c95d426c4edc7ae3f9d723d3ce1a7d5c3.jpg", + "image_caption": [ + "Figure 5. Qualitative comparison. The first two rows show the results on Thuman2.0 dataset while the last two rows show the results on MVHumaNet dataset. Our method achieves good texture preservation (highlighted by the blue boxes), while three baseline methods mostly fail." + ], + "image_footnote": [], + "bbox": [ + 133, + 203, + 867, + 612 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "regenerates 3D clothed humans from text prompts but struggles to accurately retain both body and clothing characteristics. GaussCtrl [48], lacking support for image prompts, fails to maintain detailed clothing textures. While TipEditor [63] leverages Lora [21] for personalization, it encounters difficulties in consistently mapping clothing inputs from two views into the 3D human because the personalized concept are semantic in 2D space. In contrast, our method effectively preserves intricate clothing details, such as text, stripes, and logos.", + "bbox": [ + 88, + 678, + 480, + 829 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Quantitative Evaluation. Tab. 1 shows the results for the CLIP Directional Consistency Score and DINO similarity on Thuman2.0 and MVHumanNet datasets. Our approach surpasses other methods on $\\mathrm{DINO}_{sim}$ , clearly illustrating", + "bbox": [ + 88, + 839, + 482, + 902 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "the superiority of our method in terms of garment texture preservation. While our results on $\\mathrm{CLIP}_{\\text {cons }}$ are comparable to those of other methods, it is important to note that these methods incorporate the SDS loss, which to some extent smoothes the representation of humans in 3D space. Additionally, the \"flatter\" textures of other methods could also result in artificially higher consistency scores. Furthermore, user studies have shown that our method significantly exceeds baselines in terms of edited 3D human quality and the alignment of clothing details.", + "bbox": [ + 511, + 678, + 906, + 829 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5.3. Visual Results using E-commerce Garment", + "text_level": 1, + "bbox": [ + 511, + 845, + 879, + 862 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Fig. 6 showcases VTON results using garments from the MVG dataset [46], whose images are from e-commerce", + "bbox": [ + 511, + 869, + 906, + 902 + ], + "page_idx": 6 + }, + { + "type": "page_number", + "text": "26394", + "bbox": [ + 478, + 944, + 519, + 957 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/7737d81fbc34085b8a538160848be10370bc9d5255819ab87e19ccf3b3907f3f.jpg", + "image_caption": [ + "Figure 6. Generalization to e-commerce garments (the MVG dataset). Our method, trained on the THuman2.0 dataset, shows strong generalizability when applied to e-commerce garments. For clarity in visualization, we display garment images on human models; however, in the actual VTON process, the garments are segmented from the models using parse maps." + ], + "image_footnote": [], + "bbox": [ + 94, + 89, + 480, + 340 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "platforms like YOOX NET-A-PORTER, Taobao, and TikTok*, and a model trained on the Thuman2.0 dataset [55]. The results demonstrate that our method effectively preserves intricate garment details and textures. For instance, it accurately maintains the stripe patterns in the first row, the cute tie in the second row, and the buttons in the third row, highlighting the robustness of our approach in handling diverse and realistic clothing items.", + "bbox": [ + 89, + 455, + 483, + 575 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5.4. Ablation Study", + "text_level": 1, + "bbox": [ + 89, + 590, + 243, + 606 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We conduct an ablation study on Thuman2.0 dataset in Tab. 2 and Fig. 7 to evaluate the impact of our three proposed modules in enhancing a typical 2D VTON network with 3D-consistent generation capabilities. Starting with the 2D VTON baseline [53] using DensePose, we progressively replace DensePose with our pseudo-3D pose, incorporate multi-view CLIP embeddings, and ultimately integrate MVAttention in the final configuration. Results in Tab. 2 indicate that each module contributes to metric improvements. Fig. 7 visualizes an example of multi-view image editing. The incorporation of pseudo-3D pose substantially improves limb generation compared to the 2D VTON baseline. Comparing rows 4 and 5, prior to the integration of multi-view CLIP embedding, the model captures limited spatial information, resulting in detail loss at specific angles (columns 3, 4, and 6). Finally, the proposed MVAttention achieves a more coherent generation across views.", + "bbox": [ + 89, + 614, + 483, + 869 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/5de364b5b01530170b9e9a96febaa852e03b779b915dc80ff95da76bb7a36c53.jpg", + "image_caption": [ + "Figure 7. Visualization of the impact of the three proposed techniques on multi-view consistent editing. The red boxes highlight the artifacts. Starting from the 2D VTON baseline, the pseudo-3D pose improves limb generation quality, multi-view CLIP embedding enhances detail across different viewing directions, and finally, MVAttention further strengthens consistency in the generated images." + ], + "image_footnote": [], + "bbox": [ + 535, + 88, + 883, + 412 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/f705e5297b3e2f00053ee946d9e1e39d2ea67a67e62a277374d728081f02ef49.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
MethodsCLIPcons↑DINOsim↑
2D-VTON0.8920.609
+ Pseudo-3D Pose0.9100.626
+ Multi-view CLIP Embedding0.9130.631
+ Multi-view Spatial Attention0.9230.633
", + "bbox": [ + 517, + 526, + 898, + 619 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Table 2. Ablation studies. We ablate the impact of the three proposed techniques on Thuman2.0 dataset.", + "bbox": [ + 511, + 623, + 905, + 651 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "6. Conclusions", + "text_level": 1, + "bbox": [ + 513, + 672, + 640, + 689 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In this work, we proposed VTON 360, a novel 3D Virtual Try-On (VTON) method that achieves high-fidelity VTON with the ability to render clothing from arbitrary viewing directions. Our method features a novel formulation of 3D VTON as an extension of 2D VTON that ensures 3D consistent results across multiple views. To bridge the gap between 2D VTON models and 3D consistency requirements, we introduce several key innovations, including multi-view inputs, pseudo-3D pose representation, multi-view spatial attention, and multi-view CLIP embedding. Extensive experiments demonstrate the effectiveness of our approach, significantly outperforming prior 3D VTON techniques in both fidelity and any-view rendering.", + "bbox": [ + 511, + 703, + 906, + 900 + ], + "page_idx": 7 + }, + { + "type": "page_footnote", + "text": "* https://net-a-porter.com, www.taobao.com, www.douyin.com", + "bbox": [ + 109, + 886, + 439, + 900 + ], + "page_idx": 7 + }, + { + "type": "page_number", + "text": "26395", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Acknowledgement", + "text_level": 1, + "bbox": [ + 91, + 90, + 250, + 107 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "This work is supported in part by the National Key R&D Program of China under Grant No.2024YFB3908503, in part by the National Natural Science Foundation of China under Grant NO. 62322608 and in part by the CCF-Kuaishou Large Model Explorer Fund (NO. CCF-KuaiShou 2024007).", + "bbox": [ + 89, + 114, + 483, + 205 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "References", + "text_level": 1, + "bbox": [ + 91, + 220, + 187, + 237 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[1] Easymocap - make human motion capture easier. Github, 2021.5", + "[2] Shuai Bai, Huiling Zhou, Zhikang Li, Chang Zhou, and Hongxia Yang. Single stage virtual try-on via deformable attention flows. In European Conference on Computer Vision, pages 409-425. Springer, 2022. 2", + "[3] Bharat Lal Bhatnagar, Garvita Tiwari, Christian Theobalt, and Gerard Pons-Moll. Multi-garment net: Learning to dress 3d people from images. In Proceedings of the IEEE/CVF international conference on computer vision, pages 5420-5430, 2019. 2, 3", + "[4] Robert Bridson, Ronald Fedkiw, and John Anderson. Robust treatment of collisions, contact and friction for cloth animation. In Proceedings of the 29th annual conference on Computer graphics and interactive techniques, pages 594-603, 2002. 3", + "[5] Tim Brooks, Aleksander Holynski, and Alexei A Efros. Instructpix2pix: Learning to follow image editing instructions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18392-18402, 2023. 3", + "[6] Haodong Chen, Yongle Huang, Haojian Huang, Xiangsheng Ge, and Dian Shao. Gaussianvton: 3d human virtual try-on via multi-stage gaussian splatting editing with image prompting. arXiv preprint arXiv:2405.07472, 2024. 2, 3, 1", + "[7] Yiwen Chen, Zilong Chen, Chi Zhang, Feng Wang, Xiaofeng Yang, Yikai Wang, Zhongang Cai, Lei Yang, Huaping Liu, and Guosheng Lin. Gaussianeditor: Swift and controllable 3d editing with gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21476-21485, 2024. 3", + "[8] Seunghwan Choi, Sunghyun Park, Minsoo Lee, and Jaegul Choo. Viton-hd: High-resolution virtual try-on via misalignment-aware normalization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14131-14140, 2021. 1, 2", + "[9] Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for virtual try-on. arXiv preprint arXiv:2403.05139, 2024. 3, 4, 5", + "[10] Xin Dong, Fuwei Zhao, Zhenyu Xie, Xijin Zhang, Daniel K Du, Min Zheng, Xiang Long, Xiaodan Liang, and Jianchao Yang. Dressing in the wild by watching dance videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3480-3489, 2022. 2" + ], + "bbox": [ + 93, + 247, + 483, + 900 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[11] Yuying Ge, Yibing Song, Ruimao Zhang, Chongjian Ge, Wei Liu, and Ping Luo. Parser-free virtual try-on via distilling appearance flows. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8485-8493, 2021. 2", + "[12] Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang. Taming the power of diffusion models for high-quality virtual try-on with appearance flow. arXiv preprint arXiv:2308.06101, 2023. 1, 2, 4", + "[13] Peng Guan, Loretta Reiss, David A Hirshberg, Alexander Weiss, and Michael J Black. Drape: Dressing any person. ACM Transactions on Graphics (ToG), 31(4):1-10, 2012. 1, 3", + "[14] Riza Alp Güler, Natalia Neverova, and Iasonas Kokkinos. Densepose: Dense human pose estimation in the wild. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7297-7306, 2018. 4, 5", + "[15] Fabian Hahn, Bernhard Thomaszewski, Stelian Coros, Robert W Sumner, Forrester Cole, Mark Meyer, Tony DeRose, and Markus Gross. Subspace clothing simulation using adaptive bases. ACM Transactions on Graphics (TOG), 33(4):1-9, 2014. 1, 3", + "[16] Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S Davis. Viton: An image-based virtual try-on network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7543-7552, 2018. 2", + "[17] Ayaan Haque, Matthew Tancik, Alexei A Efros, Aleksander Holynski, and Angjoo Kanazawa. Instruct-nerf2nerf: Editing 3d scenes with instructions. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 19740–19750, 2023. 3, 6, 1", + "[18] Sen He, Yi-Zhe Song, and Tao Xiang. Style-based global appearance flow for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3470-3479, 2022. 2", + "[19] Zijian He, Peixin Chen, Guangrun Wang, Guanbin Li, Philip HS Torr, and Liang Lin. Wildvidfit: Video virtual try-on in the wild via image-based controlled diffusion models. In European Conference on Computer Vision, pages 123-139. Springer, 2024. 1, 2", + "[20] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in Neural Information Processing Systems, 33:6840-6851, 2020. 3", + "[21] Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021. 2, 3, 7", + "[22] Li Hu, Xin Gao, Peng Zhang, Ke Sun, Bang Zhang, and Liefeng Bo. Animate anyone: Consistent and controllable image-to-video synthesis for character animation. arXiv preprint arXiv:2311.17117, 2023. 5", + "[23] Yukun Huang, Jianan Wang, Ailing Zeng, He Cao, Xianbiao Qi, Yukai Shi, Zheng-Jun Zha, and Lei Zhang. Dreamwaltz: Make a scene with complex 3d animatable avatars. Advances in Neural Information Processing Systems, 36, 2024. 3, 6, 7", + "[24] Yangyi Huang, Hongwei Yi, Yuliang Xiu, Tingting Liao, Ji-axiang Tang, Deng Cai, and Justus Thies. Tech: Text-guided" + ], + "bbox": [ + 516, + 92, + 903, + 900 + ], + "page_idx": 8 + }, + { + "type": "page_number", + "text": "26396", + "bbox": [ + 478, + 944, + 519, + 955 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "reconstruction of lifelike clothed humans. In 2024 International Conference on 3D Vision (3DV), pages 1531-1542. IEEE, 2024. 3", + "[25] Bernhard Kerbl, Georgios Kopanas, Thomas Leimkuhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139-1, 2023. 3, 1", + "[26] Jeongho Kim, Gyojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on. arXiv preprint arXiv:2312.01725, 2023. 3, 4, 5", + "[27] Diederik P Kingma. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013. 3", + "[28] Zorah Lahner, Daniel Cremers, and Tony Tung. Deepwrinkles: Accurate and realistic clothing modeling. In Proceedings of the European conference on computer vision (ECCV), pages 667-684, 2018. 1, 3", + "[29] Sangyun Lee, Gyojung Gu, Sunghyun Park, Seunghwan Choi, and Jaegul Choo. High-resolution virtual try-on with misalignment and occlusion-handled conditions. In Proceedings of the European conference on computer vision (ECCV), 2022. 2", + "[30] Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J Black. Smpl: A skinned multiperson linear model. In Seminal Graphics Papers: Pushing the Boundaries, Volume 2, pages 851-866. 2023. 3", + "[31] Yifang Men, Yiming Mao, Yuning Jiang, Wei-Ying Ma, and Zhouhui Lian. Controllable person image synthesis with attribute-decomposed gan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5084-5093, 2020. 2", + "[32] Aymen Mir, Thiemo Alldieck, and Gerard Pons-Moll. Learning to transfer texture from clothing images to 3d humans. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7023-7034, 2020. 2, 3", + "[33] Davide Morelli, Alberto Baldrati, Giuseppe Cartella, Marcelia Cornia, Marco Bertini, and Rita Cucchiara. Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on. arXiv preprint arXiv:2305.13501, 2023. 1, 2", + "[34] Maxime Oquab, Timothee Darcet, Theo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 6", + "[35] Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed AA Osman, Dimitrios Tzionas, and Michael J Black. Expressive body capture: 3d hands, face, and body from a single image. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10975-10985, 2019. 5", + "[36] Gerard Pons-Moll, Sergi Pujades, Sonny Hu, and Michael J Black. Clothcap: Seamless 4d clothing capture and retargeting. ACM Transactions on Graphics (ToG), 36(4):1-15, 2017. 3", + "[37] Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Mildenhall. Dreamfusion: Text-to-3d using 2d diffusion. arXiv preprint arXiv:2209.14988, 2022. 2, 3" + ], + "bbox": [ + 91, + 90, + 482, + 900 + ], + "page_idx": 9 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[38] Yurui Ren, Xiaoqing Fan, Ge Li, Shan Liu, and Thomas H Li. Neural texture extraction and distribution for controllable person image synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13535-13544, 2022. 2", + "[39] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10684-10695, 2022. 1, 3, 4, 6", + "[40] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. arXiv preprint arXiv:2208.12242, 2022. 3", + "[41] Igor Santesteban, Nils Thuerey, Miguel A Otaduy, and Dan Casas. Self-supervised collision handling via generative 3d garment models for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11763-11773, 2021. 2", + "[42] Igor Santesteban, Miguel Otaduy, Nils Thuerey, and Dan Casas. Ulnef: Untangled layered neural fields for mix-and-match virtual try-on. Advances in Neural Information Processing Systems, 35:12110-12125, 2022. 2", + "[43] Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International Conference on Machine Learning, pages 2256-2265. PMLR, 2015. 3", + "[44] Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020.3", + "[45] Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. Advances in Neural Information Processing Systems, 32, 2019. 3", + "[46] Haoyu Wang, Zhilu Zhang, Donglin Di, Shiliang Zhang, and Wangmeng Zuo. Mv-vton: Multi-view virtual try-on with diffusion models. arXiv preprint arXiv:2404.17364, 2024. 7", + "[47] Junjie Wang, Jiemin Fang, Xiaopeng Zhang, Lingxi Xie, and Qi Tian. Gaussianeditor: Editing 3d gaussians delicately with text instructions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20902-20911, 2024. 3", + "[48] Jing Wu, Jia-Wang Bian, Xinghui Li, Guangrun Wang, Ian Reid, Philip Torr, and Victor Adrian Prisacariu. Gaussctrl: multi-view consistent text-driven 3d gaussian splattering editing. arXiv preprint arXiv:2403.08733, 2024. 3, 6, 7", + "[49] Jay Zhangjie Wu, Yixiao Ge, Xintao Wang, Stan Weixian Lei, Yuchao Gu, Yufei Shi, Wynne Hsu, Ying Shan, Xiaohu Qie, and Mike Zheng Shou. Tune-a-video: One-shot tuning of image diffusion models for text-to-video generation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 7623-7633, 2023. 5", + "[50] Zhenyu Xie, Haoye Dong, Yufei Gao, Zehua Ma, and Xiaodan Liang. Dreamvton: Customizing 3d virtual try-on with personalized diffusion models. arXiv preprint arXiv:2407.16511, 2024. 2, 3", + "[51] Zhangyang Xiong, Chenghong Li, Kenkun Liu, Hongjie Liao, Jianqiao Hu, Junyi Zhu, Shuliang Ning, Lingteng Qiu," + ], + "bbox": [ + 516, + 90, + 906, + 900 + ], + "page_idx": 9 + }, + { + "type": "page_number", + "text": "26397", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 9 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "Chongjie Wang, Shijie Wang, et al. Mvhumannet: A large-scale dataset of multi-view daily dressing human captures. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19801-19811, 2024. 2, 6, 7", + "[52] Dejia Xu, Weili Nie, Chao Liu, Sifei Liu, Jan Kautz, Zhangyang Wang, and Arash Vahdat. Camco: Camera-controllable 3d-consistent image-to-video generation. arXiv preprint arXiv:2406.02509, 2024. 5", + "[53] Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen. Oot-diffusion: Outfitting fusion based latent diffusion for controllable virtual try-on. arXiv preprint arXiv:2403.01779, 2024. 3, 4, 8", + "[54] Han Yang, Xinrui Yu, and Ziwei Liu. Full-range virtual try-on with recurrent tri-level transform. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3460-3469, 2022. 2", + "[55] Tao Yu, Zerong Zheng, Kaiwen Guo, Pengpeng Liu, Qionghai Dai, and Yebin Liu. Function4d: Real-time human volumetric capture from very sparse consumer rgbd sensors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5746-5756, 2021. 2, 6, 7, 8", + "[56] Jinsong Zhang, Kun Li, Yu-Kun Lai, and Jingyu Yang. Pise: Person image synthesis and editing with decoupled gan. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7982-7990, 2021. 2", + "[57] Lvmin Zhang and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. arXiv preprint arXiv:2302.05543, 2023. 3", + "[58] Yabo Zhang, Yuxiang Wei, Dongsheng Jiang, Xiaopeng Zhang, Wangmeng Zuo, and Qi Tian. Controlvideo: Training-free controllable text-to-video generation. arXiv preprint arXiv:2305.13077, 2023. 5", + "[59] Fuwei Zhao, Zhenyu Xie, Michael Kampffmeyer, Haoye Dong, Songfang Han, Tianxiang Zheng, Tao Zhang, and Xiaodan Liang. M3d-vton: A monocular-to-3d virtual try-on network. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 13239-13249, 2021. 3", + "[60] Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Sahara, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two unets. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4606-4615, 2023. 2", + "[61] Jingyu Zhuang, Chen Wang, Liang Lin, Lingjie Liu, and Guanbin Li. Dreameditor: Text-driven 3d scene editing with neural fields. In SIGGRAPH Asia 2023 Conference Papers, pages 1-10, 2023. 3", + "[62] Jingyu Zhuang, Di Kang, Linchao Bao, Liang Lin, and Guanbin Li. Dagensm: Disentangled avatar generation with gs-enhanced mesh. arXiv preprint arXiv:2411.15205, 2024.3", + "[63] Jingyu Zhuang, Di Kang, Yan-Pei Cao, Guanbin Li, Liang Lin, and Ying Shan. Tip-editor: An accurate 3d editor following both text-prompts and image-prompts. arXiv preprint arXiv:2401.14828, 2024. 3, 6, 7, 1" + ], + "bbox": [ + 91, + 90, + 482, + 898 + ], + "page_idx": 10 + }, + { + "type": "page_number", + "text": "26398", + "bbox": [ + 478, + 945, + 517, + 955 + ], + "page_idx": 10 + } +] \ No newline at end of file diff --git a/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/4af7cfba-2431-4854-85f1-f3f911dcc811_model.json b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/4af7cfba-2431-4854-85f1-f3f911dcc811_model.json new file mode 100644 index 0000000000000000000000000000000000000000..caa4d38d60167ef1b6b477b48ddc2905d8759fac --- /dev/null +++ b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/4af7cfba-2431-4854-85f1-f3f911dcc811_model.json @@ -0,0 +1,2103 @@ +[ + [ + { + "type": "header", + "bbox": [ + 0.107, + 0.003, + 0.182, + 0.043 + ], + "angle": 0, + "content": "CVF" + }, + { + "type": "header", + "bbox": [ + 0.237, + 0.001, + 0.808, + 0.047 + ], + "angle": 0, + "content": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore." + }, + { + "type": "title", + "bbox": [ + 0.143, + 0.131, + 0.854, + 0.153 + ], + "angle": 0, + "content": "VTON 360: High-Fidelity Virtual Try-On from Any Viewing Direction" + }, + { + "type": "text", + "bbox": [ + 0.1, + 0.16, + 0.9, + 0.251 + ], + "angle": 0, + "content": "Zijian He1 Yuwei Ning1 Yipeng Qin2 Guangrun Wang1 Sibei Yang3 Liang Lin1,4,5 Guanbin Li1,4,5* \n1Sun Yat-sen University \n2Cardiff University \n3ShanghaiTech University \n4Guangdong Key Laboratory of Big Data Analysis and Processing \n5Peng Cheng Laboratory \nhezj39@mail2.sysu.edu.cn, yuwei.ning@hust.edu.cn, {qinyipengl991, wanggrun}@gmail.com, yangsb@shanghaiitech.edu.cn linliang@ieee.org, liguanbin@mail.sysu.edu.cn" + }, + { + "type": "image", + "bbox": [ + 0.092, + 0.253, + 0.511, + 0.475 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.518, + 0.253, + 0.71, + 0.476 + ], + "angle": 0, + "content": null + }, + { + "type": "image", + "bbox": [ + 0.715, + 0.253, + 0.907, + 0.476 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.089, + 0.481, + 0.908, + 0.54 + ], + "angle": 0, + "content": "Figure 1. Results of VTON 360. Our VTON 360 enables high-fidelity 3D Virtual Try-On (VTON) by seamlessly adapting E-commerce garments onto a clothed 3D human model, supporting full \\(360^{\\circ}\\) view rendering. The highlighted bounding boxes (dashed line) demonstrate our method's ability to preserve intricate clothing details and patterns (e.g., collar accessories, horizontal line patterns, logos, texts, numbers) across diverse garment types." + }, + { + "type": "title", + "bbox": [ + 0.248, + 0.544, + 0.327, + 0.559 + ], + "angle": 0, + "content": "Abstract" + }, + { + "type": "text", + "bbox": [ + 0.089, + 0.57, + 0.486, + 0.873 + ], + "angle": 0, + "content": "Virtual Try-On (VTON) is a transformative technology in e-commerce and fashion design, enabling realistic digital visualization of clothing on individuals. In this work, we propose VTON 360, a novel 3D VTON method that addresses the open challenge of achieving high-fidelity VTON that supports any-view rendering. Specifically, we leverage the equivalence between a 3D model and its rendered multiview 2D images, and reformulate 3D VTON as an extension of 2D VTON that ensures 3D consistent results across multiple views. To achieve this, we extend 2D VTON models to include multi-view garments and clothing-agnostic human body images as input, and propose several novel techniques to enhance them, including: i) a pseudo-3D pose representation using normal maps derived from the SMPL-X 3D human model, ii) a multi-view spatial attention mechanism that models the correlations between features from different viewing angles, and iii) a multi-view CLIP embedding that enhances the garment CLIP features used in 2D VTON with camera information. Extensive experiments on large-scale real datasets and clothing images from e-commerce" + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.545, + 0.907, + 0.59 + ], + "angle": 0, + "content": "platforms demonstrate the effectiveness of our approach. \nProject page: https://scnuhealthy.github.io/ \nVTON360." + }, + { + "type": "title", + "bbox": [ + 0.514, + 0.602, + 0.645, + 0.617 + ], + "angle": 0, + "content": "1. Introduction" + }, + { + "type": "text", + "bbox": [ + 0.511, + 0.628, + 0.907, + 0.749 + ], + "angle": 0, + "content": "Virtual Try-On (VTON) enables realistic digital visualization of clothing on individuals and has emerged as a transformative technology in e-commerce and fashion design. While significant research efforts have been made on 2D VTON solutions [8, 12, 19, 33, 39], these approaches are inherently limited in their representation of view-related features. To overcome this limitation and enable high-fidelity any-view rendering, 3D VTON methods were introduced." + }, + { + "type": "text", + "bbox": [ + 0.511, + 0.75, + 0.909, + 0.902 + ], + "angle": 0, + "content": "3D VTON requires accurate garment transfer onto a 3D human body while ensuring realistic garment fitting, texture preservation, and 3D consistency. The two primary aims of 3D VTON are i) achieving high-fidelity and ii) supporting any-view rendering. Leveraging the inherent capability of 3D models for any-view rendering, early 3D VTON methods [13, 15, 28] make clothing simulation on synthetic human bodies. Specifically, these methods utilized 3D scanners to capture clothing meshes, followed by the development of specialized dressing algorithms. Although effec" + }, + { + "type": "page_footnote", + "bbox": [ + 0.11, + 0.887, + 0.312, + 0.9 + ], + "angle": 0, + "content": "* Corresponding author is Guanbin Li." + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "26388" + } + ], + [ + { + "type": "text", + "bbox": [ + 0.09, + 0.092, + 0.482, + 0.304 + ], + "angle": 0, + "content": "tive, these methods rely on costly 3D scanning equipment and the physical presence of the human body/clothing (i.e., not fully virtual), restricting their practicality in real-world applications. As a byproduct, most early methods focused on developing geometrically correct dressing algorithms using standard templates of human body and clothing models. Addressing this limitation, researchers extended 3D VTON by introducing algorithms that reconstruct 3D clothing models from input images, enabling the use of image-based clothing inputs [3, 32, 41, 42]. However, since input clothing images (usually frontal) are inherently 2D and lack multi-view information, this approach struggles to reconstruct high-fidelity clothing models that can be rendered well from all viewing directions." + }, + { + "type": "text", + "bbox": [ + 0.093, + 0.308, + 0.483, + 0.716 + ], + "angle": 0, + "content": "To complement this missing information, DreamVTON [50] introduces a novel approach that leverages Text-to-Image (T2I) diffusion models to reconstruct both the human body and clothing from input images. Its key insight is that T2I models learned view-agnostic \"concepts\" of both bodies and garments during their training, and that the corresponding concepts for the input body and clothing images can be obtained using LoRA [21]. By utilizing Score Distillation Sampling (SDS) [37], DreamVTON can generate visual-pleasing 3D VTON results by ensuring consistency between renderings from arbitrary viewpoints and the concepts. Nonetheless, DreamVTON's high flexibility comes at the cost of low fidelity. This limitation stems from the fact that the concepts learned by T2I models are semantic in nature, thus lacking 3D geometric consistency and pixel-level accuracy with respect to the input body and clothing images. Recently, a concurrent work, namely GaussianVTON [6], partially addressed this limitation by formulating 3D VTON as a 3D scene editing task, where a given 3D human model is edited using multi-view images generated by 2D VTON methods. While it significantly enhances the fidelity of the human body, the fidelity and 3D consistency of clothing remain problematic, as there are no 2D VTON methods that can generate multi-view images with 3D consistency. Therefore, to the best of our knowledge, achieving high-fidelity 3D VTON that supports any-view rendering remains an open challenge." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.72, + 0.483, + 0.902 + ], + "angle": 0, + "content": "In this work, we address the above-mentioned challenge via proposing VTON 360, a novel 3D VTON method that achieves high-fidelity VTON from arbitrary viewing directions. Similar to GaussianVTON [6], our method edits a given 3D human model by inpainting the rendered images using a latent diffusion model. However, we set ourselves apart through our novel garment fidelity preservation strategy that can generate high-fidelity on-body garments in all viewing directions. Specifically, we first extend both the garment and clothing-agnostic human body inputs to typical 2D VTON models to leverage multi-view information, including paired front and back view garment images as well" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.092, + 0.905, + 0.483 + ], + "angle": 0, + "content": "as a set of multi-view clothing-agnostic human body images sampled from random azimuth angles. Then, we propose several novel enhancements to bridge the gap between typical 2D VTON methods and our multi-view 3D consistency requirements: i) We propose a pseudo-3D pose representation using normal maps derived from the SMPL-X 3D human model, which captures fine-grained surface orientation details and provides more consistent geometry across views compared to the 2D pose representations (semantic segmentation maps) used in 2D VTON models. ii) We design a Multi-view Spatial Attention mechanism that models the correlations between features from different viewing angles, featuring a novel \"correlation\" matrix modeling the relationships among different input views. iii) We propose a multi-view CLIP embedding that enhances the garment CLIP embedding used in 2D VTON methods with camera information, thereby facilitating network learning of features relevant to a particular view. Together, these innovations enable our 2D VTON model to generate high-quality, multi-view and 3D-consistent virtual try-on results. Extensive experiments on Thuman2.0 [55] and MVHumanNet [51] datasets demonstrate that our method achieves high fidelity 3D VTON which supports any-view rendering. In addition, we show the effectiveness and generalizability of our methodology by testing it using garments from e-commerce platforms. Our conclusions include:" + }, + { + "type": "text", + "bbox": [ + 0.514, + 0.484, + 0.905, + 0.529 + ], + "angle": 0, + "content": "- We propose a novel 3D Virtual Try-On (VTON) method, namely VTON 360, which achieves high-fidelity VTON from arbitrary viewing directions." + }, + { + "type": "text", + "bbox": [ + 0.514, + 0.53, + 0.905, + 0.664 + ], + "angle": 0, + "content": "- Leveraging the equivalence between a 3D model and its rendered multi-view 2D images, we reformulate 3D VTON as an extension of 2D VTON that ensures consistent results across multiple views. Specifically, we introduce several novel techniques, including: (i) pseudo-3D pose representation; (ii) multi-view spatial attention; and (iii) multi-view CLIP embedding. These innovations enhance traditional 2D VTON models to generate multiview and 3D-consistent results." + }, + { + "type": "text", + "bbox": [ + 0.514, + 0.665, + 0.904, + 0.711 + ], + "angle": 0, + "content": "- Extensive experimental results on two large real datasets as well as real clothing images from e-commerce platforms demonstrate the effectiveness of our approach." + }, + { + "type": "list", + "bbox": [ + 0.514, + 0.484, + 0.905, + 0.711 + ], + "angle": 0, + "content": null + }, + { + "type": "title", + "bbox": [ + 0.514, + 0.724, + 0.655, + 0.74 + ], + "angle": 0, + "content": "2. Related Work" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.75, + 0.905, + 0.901 + ], + "angle": 0, + "content": "2D Virtual Try-On. 2D Virtual Try-On (VTON) aims to transfer a desired garment to the corresponding region of a target human image while preserving the human pose and identity. Early methods [2, 8, 10, 11, 16, 18, 29, 31, 38, 54, 56] use Generative Adversarial Networks (GANs) to deform the garments to match the target body shape, which a critical step for achieving realistic VTON. However, accurately adapting to diverse real-world conditions remains a significant challenge. Addressing this issue, recent methods [12, 19, 33, 60] reframe 2D VTON as a conditioned in-" + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "26389" + } + ], + [ + { + "type": "text", + "bbox": [ + 0.09, + 0.092, + 0.482, + 0.168 + ], + "angle": 0, + "content": "painting task, leveraging the strong priors provided by diffusion models [20, 43, 45] to achieve promising results. This strategy is further improved by [9, 26, 53], which introduce a ReferenceNet to extract hierarchical garment features and apply attention mechanisms to condition the Main UNet." + }, + { + "type": "text", + "bbox": [ + 0.093, + 0.179, + 0.485, + 0.709 + ], + "angle": 0, + "content": "3D Virutal Try-On. For 3D Virtual Try-On (VTON), traditional methods [4, 13, 15, 28, 36] rely on 3D scanning or cloth simulation to generate highly precise body and garment geometry. These methods were then extended by learning-based methods [3, 32] that employ differentiable rendering to dress the SMPL [30] model with a desired garment mesh. Despite their effectiveness, such methods rely on costly 3D scanning and the physical presence of human body/clothing, limiting their application in the real world. Addressing this limitation, M3D-VTON [59] proposes a depth-based 3D VTON framework to reconstruct 3D clothed human models from 2D human and garment images, but the results often suffer from explicit warping artifacts. To improve 3D VTON results, recent methods [23, 24, 50, 62] resort to text-to-image (T2I) diffusion models and employ the Score Distillation Sampling (SDS) loss [40] to ensure consistency among different viewing directions. Specifically, TeCH [24] adapts the generative priors of T2I diffusion model to the specific person and clothes by training descriptive text prompts with DreamBooth [40]. DreamWaltz [23] leverages Pose ControlNet [57] to attain clothed human body models. DreamVTON [50] introduces a multi-concept LoRA [21] to personalize the T2I diffusion model, and uses a template-based optimization mechanism that combines with SDS loss to better preserve patterns on the garment. Although effective, these methods often produce results lacking in fidelity, as the concepts learned by T2I models are semantic rather than at the pixel level. Concurrent to our work, GaussianVTON [6] proposes an alternative approach by combining Gaussian Splatting [25] with pre-trained 2D VTON models and formulate it as an editing task. However, since there are no 2D VTON methods that can generate multi-view images with 3D consistency, the fidelity and 3D consistency of the clothing generated remain problematic." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.72, + 0.483, + 0.901 + ], + "angle": 0, + "content": "Radiance Field-based 3D Human or Scene Editing. Recently, radiance field-based editing has attracted significant interest due to its efficient differentiable rendering capabilities, sparking substantial advancements in text-driven 3D editing. For example, InstructN2N [17] employ an image-based diffusion model InstructP2P [5] to modify the rendered image by the user's text description with a variant of the score distillation sampling (SDS) [37] loss. GaussianEditor [7] applies Gaussian Splatting [25] as 3D representation instead of NeRF, adopting Gaussian semantic tracking to track target Gaussian values, significantly improving editing speed and controllability. To enable accu" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.092, + 0.908, + 0.289 + ], + "angle": 0, + "content": "rate location and appearance control, subsequent works [47, 61] specify the editing region using the attention score or with a segmentation model. TIP-Editor [63] proposes a content personalization step dedicated to the reference image based on LoRA, achieving the editing following a hybrid text-image prompt. GaussCtrl [48] leverage depth conditions and attention-based latent code alignment to achieve 3D-aware multi-view consistent editing instead of iteratively editing single views using SDS loss. However, these works primarily focus on global appearance modifications within a text-driven pipeline, while our approach emphasizes preserving fine textural details from different viewing directions throughout the editing process." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.301, + 0.64, + 0.318 + ], + "angle": 0, + "content": "3. Preliminary" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.326, + 0.907, + 0.448 + ], + "angle": 0, + "content": "Latent Diffusion Model. Latent Diffusion Model [39] is a variant of diffusion models that performs denoising within the latent space of a Variational Autoencoder (VAE) [27]. During training, given a fixed encoder \\(\\mathcal{E}\\), an input image \\(x\\) is transformed into its latent representation \\(z_0 = \\mathcal{E}(x)\\). A conditional diffusion model \\(\\hat{\\epsilon}_{\\theta}\\), typically implemented with a UNet architecture, is then trained using a weighted denoising score matching objective:" + }, + { + "type": "equation", + "bbox": [ + 0.584, + 0.457, + 0.907, + 0.476 + ], + "angle": 0, + "content": "\\[\n\\mathcal {L} _ {L D M} = \\mathbb {E} _ {\\mathbf {z}, \\mathbf {c}, \\boldsymbol {\\epsilon}, t} [ \\boldsymbol {\\epsilon} - \\| \\hat {\\epsilon} _ {\\theta} (\\mathbf {z} _ {\\mathbf {t}}; \\mathbf {c}, t) \\| _ {2} ^ {2} ] \\tag {1}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.485, + 0.909, + 0.592 + ], + "angle": 0, + "content": "where \\(\\mathbf{z}_t\\coloneqq \\alpha_t\\mathbf{x} + \\sigma_t\\boldsymbol {\\epsilon}\\) denotes the forward diffusion process at timestep \\(t\\) . \\(\\alpha_{t},\\sigma_{t}\\) are time-dependent functions defined by the diffusion model formulation; c denotes the conditional input and \\(\\epsilon \\sim \\mathcal{N}(\\mathbf{0},\\mathbf{1})\\) is Gaussian noise. During inference, data samples are generated by initiating from Gaussian noise \\(\\mathbf{z}_T\\sim \\mathcal{N}(\\mathbf{0},\\mathbf{1})\\) and iteratively refining it using a DDIM [44] sampler." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.604, + 0.605, + 0.619 + ], + "angle": 0, + "content": "4. Method" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.629, + 0.909, + 0.825 + ], + "angle": 0, + "content": "Our method leverages the equivalence between a 3D model and its rendered multi-view 2D images to achieve high-fidelity, any-view 3D VTON. Specifically, as Fig. 2 shows, given an input 3D human model and a garment image, our method 1) renders the 3D model into multi-view 2D images and 2) formulates 3D VTON as a consistent, unified 2D VTON process across these rendered views; 3) By reconstructing the edited images into a 3D model using existing 3D reconstruction methods, we ensure visual coherence and precise garment alignment from any viewing angle. Among them, the second step is crucial as existing 2D VTON methods [9, 26, 53] lack 3D knowledge, preventing them from generating multi-view images with 3D consistency." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.826, + 0.908, + 0.903 + ], + "angle": 0, + "content": "To address this challenge, we propose several novel techniques (Sec. 4.2) that equip a typical 2D VTON network (Sec. 4.1), which is built on a latent diffusion model [39], with the capability to generate 3D-consistent results. We use Gaussian Splatting [25] as our 3D representation." + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.521, + 0.958 + ], + "angle": 0, + "content": "26390" + } + ], + [ + { + "type": "image", + "bbox": [ + 0.091, + 0.089, + 0.905, + 0.433 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.09, + 0.442, + 0.908, + 0.498 + ], + "angle": 0, + "content": "Figure 2. Overview of VTON 360. Given an input 3D human model \\(\\mathbf{G}_{\\mathrm{src}}\\) and a pair of garment images \\((g_{f}, g_{b})\\), our method 1) renders \\(\\mathbf{G}_{\\mathrm{src}}\\) into multi-view 2D images (left) and 2) edits the rendered multi-view 2D images (middle); 3) reconstructs the edited images into a 3D model \\(\\mathbf{G}_{\\mathrm{VTON}}\\) (right). In the crucial step 2), we propose three novel techniques to equip a typical 2D VTON network with the capability to generate 3D-consistent results: 1) Pseudo-3D Pose Input, 2) Multi-view Spatial Attention, and 3) Multi-view CLIP Embedding." + }, + { + "type": "image", + "bbox": [ + 0.093, + 0.504, + 0.482, + 0.68 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.09, + 0.687, + 0.483, + 0.771 + ], + "angle": 0, + "content": "Figure 3. DensePose (2D) vs. SMPL-X normal map (pseudo-3D) representations. DensePose applies uniform labels per body part, lacking 3D consistency across views and causing artifacts and temporal inconsistencies (highlighted with red boxes). In contrast, SMPL-X normal maps capture fine surface details, ensuring geometric coherence and stable, realistic shading across views." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.785, + 0.373, + 0.801 + ], + "angle": 0, + "content": "4.1. Recap of 2D VTON Framework" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.81, + 0.484, + 0.901 + ], + "angle": 0, + "content": "Following previous works [12, 26, 53], we formulate 2D VTON as an exemplar-based inpainting problem, aiming to fill an input clothing-agnostic image \\(\\mathbf{A}\\) with a given garment image \\(g\\), where \\(\\mathbf{A}\\) is obtained following the method used in [53]. As illustrated in Fig. 2 (middle), the network architecture is based on the latent diffusion model [39] with" + }, + { + "type": "text", + "bbox": [ + 0.514, + 0.507, + 0.818, + 0.521 + ], + "angle": 0, + "content": "an encoder \\(\\mathcal{E}\\) and comprises two components:" + }, + { + "type": "text", + "bbox": [ + 0.514, + 0.521, + 0.89, + 0.536 + ], + "angle": 0, + "content": "- A GarmentNet [9, 53] that extracts features from \\(\\mathcal{E}(g)\\)." + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.536, + 0.905, + 0.641 + ], + "angle": 0, + "content": "- A Main UNet that inpaints \\(\\mathbf{A}\\) according to i) detailed garment features extracted by the GarmentNet; ii) the 2D pose of \\(\\mathbf{A}\\) represented by semantic labels using Dense-Pose [14]; iii) CLIP embeddings of input garment \\(g\\). Among them, i) and ii) together with noise are input to the self-attention layers, while iii) is input to the cross-attention layers of the Main UNet." + }, + { + "type": "list", + "bbox": [ + 0.513, + 0.521, + 0.905, + 0.641 + ], + "angle": 0, + "content": null + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.642, + 0.905, + 0.671 + ], + "angle": 0, + "content": "Both the GarmentNet and the Main UNet share the same network architecture." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.682, + 0.878, + 0.699 + ], + "angle": 0, + "content": "4.2. Multi-view 2D VTON with 3D Consistency" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.705, + 0.905, + 0.75 + ], + "angle": 0, + "content": "To enable the aforementioned 2D VTON model to generate multi-view and 3D-consistent results, we propose the following novel enhancements to its design:" + }, + { + "type": "text", + "bbox": [ + 0.514, + 0.75, + 0.888, + 0.765 + ], + "angle": 0, + "content": "Multi-view Inputs. We extend both inputs to the model:" + }, + { + "type": "text", + "bbox": [ + 0.515, + 0.766, + 0.906, + 0.886 + ], + "angle": 0, + "content": "- Multi-view Garment Inputs: We extend the input garment representation from a single image \\( g \\) to paired front and back view images \\( g_{f} \\), \\( g_{b} \\), providing comprehensive garment information across all viewing angles. Accordingly, we use the encoder \\( \\mathcal{E} \\) to encode \\( g_{f}, g_{b} \\) into their latent representations \\( \\mathcal{E}(g_{f}), \\mathcal{E}(g_{b}) \\) and feed them into GarmentNet to obtain their multi-layer features \\( F_{f}^{l} \\) and \\( F_{b}^{l} \\) at layer \\( l \\), respectively." + }, + { + "type": "text", + "bbox": [ + 0.514, + 0.886, + 0.905, + 0.902 + ], + "angle": 0, + "content": "- Multi-view Clothing-agnostic Image Inputs: Based on" + }, + { + "type": "list", + "bbox": [ + 0.514, + 0.766, + 0.906, + 0.902 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.518, + 0.957 + ], + "angle": 0, + "content": "26391" + } + ], + [ + { + "type": "text", + "bbox": [ + 0.104, + 0.092, + 0.483, + 0.211 + ], + "angle": 0, + "content": "the equivalence between a 3D human model and its rendered multi-view 2D images, we extend the input human body representation from a single, clothing-agnostic image, \\(\\mathbf{A}\\), to a set of \\(m\\) multi-view images, denoted as \\(\\mathbf{A}_1, \\mathbf{A}_2, \\dots, \\mathbf{A}_m\\). These images are sampled from random azimuth angles, allowing the 2D VTON model to access comprehensive, multi-view information from the input 3D human model." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.213, + 0.483, + 0.47 + ], + "angle": 0, + "content": "Pseudo-3D Pose Input. As shown in Fig. 3, the 2D Dense-Pose representations [14] commonly used in state-of-the-art 2D VTON methods [9, 26] assign a uniform semantic label to all pixels within each body part (e.g., thigh), inherently lack 3D geometric consistency across multiple views, and often introduce artifacts and temporal inconsistencies. To address these limitations, we propose a novel pseudo-3D pose representation: the normal maps \\(\\mathbf{N}\\) derived from the SMPL-X [35] model of the input body. These normal maps capture fine-grained surface orientation details, providing a more consistent representation across views by preserving geometric structure in the 3D space. Furthermore, they enable smoother, temporally stable transitions and realistic shading effects across multi-view scenarios. In practice, we employ a lightweight PoseEncoder \\(\\mathcal{E}'\\) [22] and feed \\(\\mathcal{E}'(\\mathbf{N})\\) into the Main UNet. We obtain the SMPL-X model from the multi-view images of the input body using EasyMoCap [1]." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.471, + 0.483, + 0.59 + ], + "angle": 0, + "content": "Accordingly, we concatenate three components as the enhanced input to the Main UNet: i) a noise latent \\( z_{t} \\); ii) the encoded pseudo-3D poses \\( \\mathcal{E}'(\\mathbf{N_1}),\\mathcal{E}'(\\mathbf{N_2}),\\dots,\\mathcal{E}'(\\mathbf{N_m}) \\); and iii) the encoded multi-view clothing-agnostic images \\( \\mathcal{E}(\\mathbf{A_1}),\\mathcal{E}(\\mathbf{A_2}),\\dots,\\mathcal{E}(\\mathbf{A_m}) \\). Let \\( F_{1}^{l},F_{2}^{l},\\dots,F_{m}^{l} \\) be the feature representations at layer \\( l \\) of the Main UNet, and recall the garment features \\( F_{f}^{l} \\) and \\( F_{b}^{l} \\) defined above, we enhance the self-attention layers of the Main UNet as:" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.592, + 0.483, + 0.818 + ], + "angle": 0, + "content": "Multi-view Spatial Attention. To cope with the aforementioned multi-view input features and ensure their consistency, we draw insights from the temporal attention layer commonly used in video generation and editing [49, 58] and extend it to our multi-view spatial attention layer, denoted as MVAttention. The key distinction of our MVAttention is that its input multi-view features \\( F_{1}^{l}, F_{2}^{l}, \\ldots, F_{m}^{l} \\) are from images captured from non-uniform spatial intervals, with the viewing angles varying randomly. Consequently, features from similar views exhibit a higher correlation, while those from distinct views are largely independent. To model this relationship, we construct a \"correlation\" matrix \\( C \\) based on the angular disparity obtained from camera rotation matrices of the input multi-view images, and define our MVAttention as follows:" + }, + { + "type": "equation", + "bbox": [ + 0.113, + 0.827, + 0.482, + 0.906 + ], + "angle": 0, + "content": "\\[\n\\begin{array}{l} \\mathbf {F} ^ {\\mathbf {l}} = \\left[ F _ {1} ^ {l} \\oplus F _ {2} ^ {l} \\dots \\oplus F _ {m} ^ {l} \\right], \\hat {\\mathbf {F}} ^ {\\mathbf {l}} = \\left[ \\mathbf {F} ^ {\\mathbf {l}} \\oplus F _ {f} ^ {l} \\oplus F _ {b} ^ {l} \\right] \\\\ Q = W ^ {Q} \\mathbf {F} ^ {1}, K = W ^ {K} \\hat {\\mathbf {F}} ^ {1}, V = W ^ {V} \\hat {\\mathbf {F}} ^ {1} \\tag {2} \\\\ A _ {i} = \\operatorname {s o f t m a x} \\left(\\frac {Q _ {i} \\times \\left(C _ {i} \\cdot K ^ {T}\\right)}{\\sqrt {d}}\\right), \\mathbf {H} _ {\\mathbf {i}} ^ {\\mathbf {l}} = A _ {i} \\times V _ {i} \\\\ \\end{array}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.092, + 0.905, + 0.228 + ], + "angle": 0, + "content": "where \\(i\\in \\{1,2,\\dots,m\\}\\) denotes \\(i\\) -th view; the Query \\(Q\\) comes directly from \\(\\mathbf{F}^1\\) and the concatenation of \\([\\mathbf{F}^1,F_f^l,F_b^l ]\\) serves as the key \\(K\\) and the value \\(V\\) . \\(\\oplus\\) indicates matrix concatenation along the token axis; \\(d\\) denotes the dimension; \\(W^{Q},W^{K},W^{V}\\) represent the linear transformation matrices; we omitted the \\(l\\) of the attention matrices and parameters for simplicity; \\(C\\in \\mathbb{R}^{m\\times m}\\) \\(C_i\\) represents \\(i\\) -th row in \\(C\\) , and its \"correlation\" value between \\(i\\) -th and \\(j\\) -th features is \\(C_{ij}\\) .." + }, + { + "type": "equation", + "bbox": [ + 0.584, + 0.24, + 0.905, + 0.259 + ], + "angle": 0, + "content": "\\[\nC _ {i j} = \\left(\\left(\\operatorname {t r a c e} \\left(R _ {i} ^ {T} R _ {j}\\right) - 1\\right) / 2 + 1\\right) / 2 \\tag {3}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.27, + 0.905, + 0.315 + ], + "angle": 0, + "content": "where \\(R_{i}\\) and \\(R_{j}\\) are the extrinsic rotation matrices of the corresponding camera views, \\((\\mathrm{trace}(R_i^T R_j) - 1) / 2\\) is the cosine value of the angle between these camera views." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.316, + 0.906, + 0.528 + ], + "angle": 0, + "content": "Multi-view CLIP Embedding. Camera viewpoints can serve as an effective condition signal to enhance 3D consistency in video content generation [52]. Building on this insight, we incorporate camera condition within our try-on network by encoding camera parameters as an additional token, enabling the generation of more consistent multi-view images. Specifically, we define a world coordinate system in which the camera faces the subject directly. For each input image (view) \\(\\mathbf{A}_{\\mathrm{i}}\\), \\(1 \\leq i \\leq m\\), we extract the rotation matrix from the camera's corresponding extrinsic matrix. This rotation matrix is then reshaped into a 9-dimensional tensor \\(\\mathbf{r}_{\\mathrm{i}}\\), which undergoes positional encoding to effectively integrate the camera parameters into the feature representation \\(F_{i}^{c}\\)." + }, + { + "type": "equation", + "bbox": [ + 0.596, + 0.54, + 0.905, + 0.579 + ], + "angle": 0, + "content": "\\[\n\\begin{array}{l} F _ {i} ^ {c} = \\left(\\sin \\left(2 ^ {0} \\pi \\mathbf {r} _ {\\mathbf {i}}\\right), \\cos \\left(2 ^ {0} \\pi \\mathbf {r} _ {\\mathbf {i}}\\right), \\dots , \\right. \\tag {4} \\\\ \\sin (2 ^ {L - 1} \\pi \\mathbf {r _ {i}}), \\cos (2 ^ {L - 1} \\pi \\mathbf {r _ {i}})) \\\\ \\end{array}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.591, + 0.906, + 0.68 + ], + "angle": 0, + "content": "where \\(L\\) is the length of positional embedding. We then project \\(F_{i}^{c}\\) to match the dimensionality of the garment CLIP image embedding \\(F^{g}\\) via an MLP and concatenate them along the token axis to form \\(Y_{i}\\). This combined representation, \\(Y_{i}\\), is subsequently used in the key \\(K_{\\mathrm{x}}\\) and value \\(V_{\\mathrm{x}}\\) of the cross-attention layers of the Main UNet:" + }, + { + "type": "equation", + "bbox": [ + 0.636, + 0.694, + 0.778, + 0.71 + ], + "angle": 0, + "content": "\\[\nY _ {i} = F ^ {g} \\oplus \\operatorname {M L P} \\left(F _ {i} ^ {c}\\right)\n\\]" + }, + { + "type": "equation", + "bbox": [ + 0.561, + 0.713, + 0.905, + 0.737 + ], + "angle": 0, + "content": "\\[\nQ _ {\\mathrm {x}} = W _ {\\mathrm {x}} ^ {Q} \\mathbf {H} _ {\\mathrm {i}} ^ {1}, K _ {\\mathrm {x}} = W _ {\\mathrm {x}} ^ {K} Y _ {i}, V _ {\\mathrm {x}} = W _ {\\mathrm {x}} ^ {V} Y _ {i} \\tag {5}\n\\]" + }, + { + "type": "equation", + "bbox": [ + 0.588, + 0.735, + 0.795, + 0.769 + ], + "angle": 0, + "content": "\\[\nF _ {i} ^ {(l + 1)} = \\mathrm {s o f t m a x} (\\frac {Q _ {\\mathrm {x}} K _ {\\mathrm {x}} ^ {T}}{\\sqrt {d _ {\\mathrm {x}}}}) V _ {\\mathrm {x}},\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.78, + 0.905, + 0.826 + ], + "angle": 0, + "content": "where \\(\\mathbf{H}^1\\) is the output of the MVAttention of the \\(l\\)-th layer; we omitted the \\(l\\) of the cross attention matrices and parameters for simplicity." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.827, + 0.905, + 0.871 + ], + "angle": 0, + "content": "Training. Our enhanced multi-view 2D VTON network can be trained by minimizing the following latent diffusion model loss function:" + }, + { + "type": "equation", + "bbox": [ + 0.555, + 0.884, + 0.905, + 0.904 + ], + "angle": 0, + "content": "\\[\n\\mathcal {L} _ {\\mathrm {l d m}} = \\mathbb {E} _ {z _ {t}, \\eta , \\psi , \\epsilon , t} \\left[ \\| \\epsilon - \\epsilon_ {\\theta} (z _ {t}, t, \\eta , \\psi , \\zeta)) \\| _ {2} ^ {2} \\right], \\tag {6}\n\\]" + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "26392" + } + ], + [ + { + "type": "image", + "bbox": [ + 0.094, + 0.089, + 0.487, + 0.295 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.09, + 0.302, + 0.487, + 0.375 + ], + "angle": 0, + "content": "Figure 4. Illustration of the proposed Multi-view Spatial Attention. Query (Q): multi-view features \\(\\mathbf{F}^1\\); Key (K) and Value (V): concatenation of \\(\\mathbf{F}^1\\) and garment features \\(F_{f}^{l}\\) and \\(F_{b}^{l}\\). The attention score between viewpoints \\(i\\) and \\(j\\) is modulated by a weight \\(C_{ij}\\), determined by the cosine of the angle between them." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.381, + 0.484, + 0.444 + ], + "angle": 0, + "content": "where \\(\\eta = [\\mathcal{E}(g_f); \\mathcal{E}(g_b); \\mathcal{E}(\\mathbf{N_i})_{i=1}^m]\\) represents the input latent garment images and latent normal maps; \\(\\zeta = [\\mathcal{E}'(\\mathbf{A_i})_{i=1}^m]\\) denotes the input latent clothing-agnostic images; \\(\\psi = \\mathbf{Y}\\) is the proposed multi-view CLIP embedding." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.457, + 0.226, + 0.475 + ], + "angle": 0, + "content": "5. Experiments" + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.482, + 0.283, + 0.5 + ], + "angle": 0, + "content": "5.1. Experimental Setup" + }, + { + "type": "text", + "bbox": [ + 0.089, + 0.504, + 0.483, + 0.776 + ], + "angle": 0, + "content": "Datasets. We conduct experiments on two public datasets: Thuman2.0 [55] and MVHumanNet [51]. Thuman2.0 comprises 526 reconstructed clothed human scans, from which we render multi-view input images. Of these samples, 426 are used for training, while the remaining 100 are set aside for testing. To further evaluate the effectiveness and robustness of our method, we also perform experiments on MVHumanNet, a large-scale dataset of multi-view human images that encompasses a diverse range of subjects, daily outfits and motion sequences. The images in MVHumanNet are captured using a multi-view system with either 48 or 24 cameras. We use 4,990 subjects from this dataset, allocating 4,790 to training and 200 for tests. For each subject, we randomly select two frames of multi-view images from its entire motion sequence. While MVHumanNet provides multi-view images directly for editing and reconstruction, we render uniformly distributed views around each human subject in Thuman2.0 to ensure consistent input." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.781, + 0.483, + 0.903 + ], + "angle": 0, + "content": "Baselines. We primarily compare our method with three existing methods: DreamWaltz [23], GaussCtrl [48], and TIP-Editor [63]. DreamWaltz is a method designed for directly generating 3D human bodies based on textual descriptions, while GaussCtrl and TIP-Editor are two radiance-based editing methods. GaussCtrl is based on Stable Diffusion, using a description-like prompt to edit the scene. TIP-Editor accepts both text and image prompts. We configure it by" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.092, + 0.905, + 0.152 + ], + "angle": 0, + "content": "specifying the human body as the editing region and the desired garment as the image prompt. We use ChatGPT to generate the text prompts corresponding to the clothing images." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.156, + 0.907, + 0.352 + ], + "angle": 0, + "content": "Evaluation Metrics. For quantitative evaluation, we assess garment-to-person alignment between the edited person and the reference image. Following [63], we calculate the average DINO similarity [34] between the reference image and the rendered multi-view images of the edited 3D scene. Additionally, to evaluate multi-view consistency, we compute the CLIP Directional Consistency Score as outlined in [17]. Given the large scale of experiments (repeated 3DGS reconstruction), we selected a subset of examples from the dataset for metric evaluation. Specifically, from the test sets of Thuman and MVHumanNet, we randomly sampled 10 human scans each, performing virtual try-on with 6 randomly chosen garments per human scan." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.352, + 0.909, + 0.489 + ], + "angle": 0, + "content": "We further conducted a user study involving 50 participants who rated the results of our method and three baseline methods based on two criteria: overall \"Quality\" and \"Alignment\" with the reference image. Each evaluation consisted of two questions: (1) Which method produces the highest quality of the edited 3D human? and (2) Which method achieves the most consistent alignment with the target clothing? Participants viewed the VTON results as rotating randomized video sequences." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.492, + 0.909, + 0.612 + ], + "angle": 0, + "content": "Implementation Details. During pre-processing, we crop the multi-view images to the bounding box around the person and resize them to a resolution of \\(768 \\times 576\\). The front view and the back view of garment images are obtained from the corresponding clothed human images. After editing, we pad the images back to their original size. The data processing pipeline is the same for both Thuman2.0 and MVHumanNet datasets." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.613, + 0.909, + 0.84 + ], + "angle": 0, + "content": "The Main UNet and the GarmentNet are initialized by the Stable Diffusion V1.5 [39]. The training process is divided into two stages. In the first stage, each view is trained independently, during which we establish the feature extraction capabilities of both the PoseEncoder and GarmentNet, as well as the generative capability of the Main UNet. The second stage involves multi-view training, where we randomly select \\( M \\) views for each human subject. This stage is focused on training the proposed MVAttention module to enhance the network's ability to maintain consistency across views. Due to memory constraints, we set \\( M = 8 \\) for the training phase. During the testing phase, we uniformly sampled 32 views from a 360-degree rotation around the subject. The editing of these 32 views is conducted in two batches, with each batch processing \\( M = 16 \\) views." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.849, + 0.89, + 0.866 + ], + "angle": 0, + "content": "5.2. Comparisons with State-of-the-Art Methods" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.871, + 0.908, + 0.903 + ], + "angle": 0, + "content": "Qualitative Evaluation. Fig. 5 shows visual comparisons between our method and the baselines. DreamWaltz [23]" + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.521, + 0.958 + ], + "angle": 0, + "content": "26393" + } + ], + [ + { + "type": "table", + "bbox": [ + 0.097, + 0.089, + 0.903, + 0.181 + ], + "angle": 0, + "content": "
MethodThuman2.0 [55]MVHumanNet [51]
CLIPcons↑DINOsim↑VotequalityVotealignCLIPcons↑DINOsim↑VotequalityVotealign
DreamWaltz [23]0.8870.5560.46%1.54%0.9350.4950.46%0.46%
TIP-Editor [63]0.9390.5690.92%0.62%0.9480.5122.15%1.38%
GaussCtrl [48]0.9310.5771.08%1.38%0.9380.5211.69%1.23%
Ours0.9230.63397.54%96.46%0.9330.62395.69%96.92%
" + }, + { + "type": "table_caption", + "bbox": [ + 0.116, + 0.186, + 0.88, + 0.201 + ], + "angle": 0, + "content": "Table 1. Quantitative comparisons. \\( {\\mathrm{{CLIP}}}_{\\text{cons }} \\) denotes the CLIP Direction Consistency Score. \\( {\\mathrm{{DINO}}}_{\\text{sim }} \\) is the DINO similarity." + }, + { + "type": "image", + "bbox": [ + 0.134, + 0.204, + 0.868, + 0.613 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.089, + 0.625, + 0.908, + 0.668 + ], + "angle": 0, + "content": "Figure 5. Qualitative comparison. The first two rows show the results on Thuman2.0 dataset while the last two rows show the results on MVHumaNet dataset. Our method achieves good texture preservation (highlighted by the blue boxes), while three baseline methods mostly fail." + }, + { + "type": "text", + "bbox": [ + 0.089, + 0.679, + 0.482, + 0.83 + ], + "angle": 0, + "content": "regenerates 3D clothed humans from text prompts but struggles to accurately retain both body and clothing characteristics. GaussCtrl [48], lacking support for image prompts, fails to maintain detailed clothing textures. While TipEditor [63] leverages Lora [21] for personalization, it encounters difficulties in consistently mapping clothing inputs from two views into the 3D human because the personalized concept are semantic in 2D space. In contrast, our method effectively preserves intricate clothing details, such as text, stripes, and logos." + }, + { + "type": "text", + "bbox": [ + 0.089, + 0.84, + 0.483, + 0.903 + ], + "angle": 0, + "content": "Quantitative Evaluation. Tab. 1 shows the results for the CLIP Directional Consistency Score and DINO similarity on Thuman2.0 and MVHumanNet datasets. Our approach surpasses other methods on \\(\\mathrm{DINO}_{sim}\\), clearly illustrating" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.679, + 0.907, + 0.83 + ], + "angle": 0, + "content": "the superiority of our method in terms of garment texture preservation. While our results on \\(\\mathrm{CLIP}_{\\text {cons }}\\) are comparable to those of other methods, it is important to note that these methods incorporate the SDS loss, which to some extent smoothes the representation of humans in 3D space. Additionally, the \"flatter\" textures of other methods could also result in artificially higher consistency scores. Furthermore, user studies have shown that our method significantly exceeds baselines in terms of edited 3D human quality and the alignment of clothing details." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.846, + 0.88, + 0.863 + ], + "angle": 0, + "content": "5.3. Visual Results using E-commerce Garment" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.871, + 0.908, + 0.903 + ], + "angle": 0, + "content": "Fig. 6 showcases VTON results using garments from the MVG dataset [46], whose images are from e-commerce" + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.521, + 0.958 + ], + "angle": 0, + "content": "26394" + } + ], + [ + { + "type": "image", + "bbox": [ + 0.095, + 0.09, + 0.482, + 0.341 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.09, + 0.361, + 0.485, + 0.446 + ], + "angle": 0, + "content": "Figure 6. Generalization to e-commerce garments (the MVG dataset). Our method, trained on the THuman2.0 dataset, shows strong generalizability when applied to e-commerce garments. For clarity in visualization, we display garment images on human models; however, in the actual VTON process, the garments are segmented from the models using parse maps." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.456, + 0.484, + 0.577 + ], + "angle": 0, + "content": "platforms like YOOX NET-A-PORTER, Taobao, and TikTok*, and a model trained on the Thuman2.0 dataset [55]. The results demonstrate that our method effectively preserves intricate garment details and textures. For instance, it accurately maintains the stripe patterns in the first row, the cute tie in the second row, and the buttons in the third row, highlighting the robustness of our approach in handling diverse and realistic clothing items." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.591, + 0.245, + 0.607 + ], + "angle": 0, + "content": "5.4. Ablation Study" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.615, + 0.484, + 0.871 + ], + "angle": 0, + "content": "We conduct an ablation study on Thuman2.0 dataset in Tab. 2 and Fig. 7 to evaluate the impact of our three proposed modules in enhancing a typical 2D VTON network with 3D-consistent generation capabilities. Starting with the 2D VTON baseline [53] using DensePose, we progressively replace DensePose with our pseudo-3D pose, incorporate multi-view CLIP embeddings, and ultimately integrate MVAttention in the final configuration. Results in Tab. 2 indicate that each module contributes to metric improvements. Fig. 7 visualizes an example of multi-view image editing. The incorporation of pseudo-3D pose substantially improves limb generation compared to the 2D VTON baseline. Comparing rows 4 and 5, prior to the integration of multi-view CLIP embedding, the model captures limited spatial information, resulting in detail loss at specific angles (columns 3, 4, and 6). Finally, the proposed MVAttention achieves a more coherent generation across views." + }, + { + "type": "image", + "bbox": [ + 0.536, + 0.089, + 0.885, + 0.413 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.513, + 0.417, + 0.908, + 0.515 + ], + "angle": 0, + "content": "Figure 7. Visualization of the impact of the three proposed techniques on multi-view consistent editing. The red boxes highlight the artifacts. Starting from the 2D VTON baseline, the pseudo-3D pose improves limb generation quality, multi-view CLIP embedding enhances detail across different viewing directions, and finally, MVAttention further strengthens consistency in the generated images." + }, + { + "type": "table", + "bbox": [ + 0.518, + 0.527, + 0.9, + 0.62 + ], + "angle": 0, + "content": "
MethodsCLIPcons↑DINOsim↑
2D-VTON0.8920.609
+ Pseudo-3D Pose0.9100.626
+ Multi-view CLIP Embedding0.9130.631
+ Multi-view Spatial Attention0.9230.633
" + }, + { + "type": "table_caption", + "bbox": [ + 0.513, + 0.624, + 0.906, + 0.652 + ], + "angle": 0, + "content": "Table 2. Ablation studies. We ablate the impact of the three proposed techniques on Thuman2.0 dataset." + }, + { + "type": "title", + "bbox": [ + 0.514, + 0.673, + 0.642, + 0.69 + ], + "angle": 0, + "content": "6. Conclusions" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.704, + 0.907, + 0.901 + ], + "angle": 0, + "content": "In this work, we proposed VTON 360, a novel 3D Virtual Try-On (VTON) method that achieves high-fidelity VTON with the ability to render clothing from arbitrary viewing directions. Our method features a novel formulation of 3D VTON as an extension of 2D VTON that ensures 3D consistent results across multiple views. To bridge the gap between 2D VTON models and 3D consistency requirements, we introduce several key innovations, including multi-view inputs, pseudo-3D pose representation, multi-view spatial attention, and multi-view CLIP embedding. Extensive experiments demonstrate the effectiveness of our approach, significantly outperforming prior 3D VTON techniques in both fidelity and any-view rendering." + }, + { + "type": "page_footnote", + "bbox": [ + 0.11, + 0.887, + 0.441, + 0.901 + ], + "angle": 0, + "content": "* https://net-a-porter.com, www.taobao.com, www.douyin.com" + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "26395" + } + ], + [ + { + "type": "title", + "bbox": [ + 0.092, + 0.091, + 0.251, + 0.108 + ], + "angle": 0, + "content": "Acknowledgement" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.116, + 0.484, + 0.207 + ], + "angle": 0, + "content": "This work is supported in part by the National Key R&D Program of China under Grant No.2024YFB3908503, in part by the National Natural Science Foundation of China under Grant NO. 62322608 and in part by the CCF-Kuaishou Large Model Explorer Fund (NO. CCF-KuaiShou 2024007)." + }, + { + "type": "title", + "bbox": [ + 0.093, + 0.222, + 0.188, + 0.238 + ], + "angle": 0, + "content": "References" + }, + { + "type": "ref_text", + "bbox": [ + 0.101, + 0.248, + 0.483, + 0.274 + ], + "angle": 0, + "content": "[1] Easymocap - make human motion capture easier. Github, 2021.5" + }, + { + "type": "ref_text", + "bbox": [ + 0.101, + 0.277, + 0.484, + 0.332 + ], + "angle": 0, + "content": "[2] Shuai Bai, Huiling Zhou, Zhikang Li, Chang Zhou, and Hongxia Yang. Single stage virtual try-on via deformable attention flows. In European Conference on Computer Vision, pages 409-425. Springer, 2022. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.335, + 0.484, + 0.403 + ], + "angle": 0, + "content": "[3] Bharat Lal Bhatnagar, Garvita Tiwari, Christian Theobalt, and Gerard Pons-Moll. Multi-garment net: Learning to dress 3d people from images. In Proceedings of the IEEE/CVF international conference on computer vision, pages 5420-5430, 2019. 2, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.406, + 0.483, + 0.473 + ], + "angle": 0, + "content": "[4] Robert Bridson, Ronald Fedkiw, and John Anderson. Robust treatment of collisions, contact and friction for cloth animation. In Proceedings of the 29th annual conference on Computer graphics and interactive techniques, pages 594-603, 2002. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.477, + 0.484, + 0.544 + ], + "angle": 0, + "content": "[5] Tim Brooks, Aleksander Holynski, and Alexei A Efros. Instructpix2pix: Learning to follow image editing instructions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18392-18402, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.548, + 0.484, + 0.615 + ], + "angle": 0, + "content": "[6] Haodong Chen, Yongle Huang, Haojian Huang, Xiangsheng Ge, and Dian Shao. Gaussianvton: 3d human virtual try-on via multi-stage gaussian splatting editing with image prompting. arXiv preprint arXiv:2405.07472, 2024. 2, 3, 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.619, + 0.484, + 0.701 + ], + "angle": 0, + "content": "[7] Yiwen Chen, Zilong Chen, Chi Zhang, Feng Wang, Xiaofeng Yang, Yikai Wang, Zhongang Cai, Lei Yang, Huaping Liu, and Guosheng Lin. Gaussianeditor: Swift and controllable 3d editing with gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21476-21485, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.704, + 0.484, + 0.773 + ], + "angle": 0, + "content": "[8] Seunghwan Choi, Sunghyun Park, Minsoo Lee, and Jaegul Choo. Viton-hd: High-resolution virtual try-on via misalignment-aware normalization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14131-14140, 2021. 1, 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.775, + 0.484, + 0.829 + ], + "angle": 0, + "content": "[9] Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for virtual try-on. arXiv preprint arXiv:2403.05139, 2024. 3, 4, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.832, + 0.483, + 0.901 + ], + "angle": 0, + "content": "[10] Xin Dong, Fuwei Zhao, Zhenyu Xie, Xijin Zhang, Daniel K Du, Min Zheng, Xiang Long, Xiaodan Liang, and Jianchao Yang. Dressing in the wild by watching dance videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3480-3489, 2022. 2" + }, + { + "type": "list", + "bbox": [ + 0.094, + 0.248, + 0.484, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.093, + 0.905, + 0.161 + ], + "angle": 0, + "content": "[11] Yuying Ge, Yibing Song, Ruimao Zhang, Chongjian Ge, Wei Liu, and Ping Luo. Parser-free virtual try-on via distilling appearance flows. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8485-8493, 2021. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.164, + 0.905, + 0.219 + ], + "angle": 0, + "content": "[12] Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang. Taming the power of diffusion models for high-quality virtual try-on with appearance flow. arXiv preprint arXiv:2308.06101, 2023. 1, 2, 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.221, + 0.905, + 0.274 + ], + "angle": 0, + "content": "[13] Peng Guan, Loretta Reiss, David A Hirshberg, Alexander Weiss, and Michael J Black. Drape: Dressing any person. ACM Transactions on Graphics (ToG), 31(4):1-10, 2012. 1, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.277, + 0.905, + 0.332 + ], + "angle": 0, + "content": "[14] Riza Alp Güler, Natalia Neverova, and Iasonas Kokkinos. Densepose: Dense human pose estimation in the wild. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7297-7306, 2018. 4, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.334, + 0.905, + 0.403 + ], + "angle": 0, + "content": "[15] Fabian Hahn, Bernhard Thomaszewski, Stelian Coros, Robert W Sumner, Forrester Cole, Mark Meyer, Tony DeRose, and Markus Gross. Subspace clothing simulation using adaptive bases. ACM Transactions on Graphics (TOG), 33(4):1-9, 2014. 1, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.406, + 0.905, + 0.46 + ], + "angle": 0, + "content": "[16] Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S Davis. Viton: An image-based virtual try-on network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7543-7552, 2018. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.462, + 0.905, + 0.531 + ], + "angle": 0, + "content": "[17] Ayaan Haque, Matthew Tancik, Alexei A Efros, Aleksander Holynski, and Angjoo Kanazawa. Instruct-nerf2nerf: Editing 3d scenes with instructions. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 19740–19750, 2023. 3, 6, 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.533, + 0.905, + 0.587 + ], + "angle": 0, + "content": "[18] Sen He, Yi-Zhe Song, and Tao Xiang. Style-based global appearance flow for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3470-3479, 2022. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.59, + 0.905, + 0.658 + ], + "angle": 0, + "content": "[19] Zijian He, Peixin Chen, Guangrun Wang, Guanbin Li, Philip HS Torr, and Liang Lin. Wildvidfit: Video virtual try-on in the wild via image-based controlled diffusion models. In European Conference on Computer Vision, pages 123-139. Springer, 2024. 1, 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.66, + 0.905, + 0.701 + ], + "angle": 0, + "content": "[20] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in Neural Information Processing Systems, 33:6840-6851, 2020. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.703, + 0.905, + 0.758 + ], + "angle": 0, + "content": "[21] Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021. 2, 3, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.76, + 0.905, + 0.815 + ], + "angle": 0, + "content": "[22] Li Hu, Xin Gao, Peng Zhang, Ke Sun, Bang Zhang, and Liefeng Bo. Animate anyone: Consistent and controllable image-to-video synthesis for character animation. arXiv preprint arXiv:2311.17117, 2023. 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.817, + 0.905, + 0.871 + ], + "angle": 0, + "content": "[23] Yukun Huang, Jianan Wang, Ailing Zeng, He Cao, Xianbiao Qi, Yukai Shi, Zheng-Jun Zha, and Lei Zhang. Dreamwaltz: Make a scene with complex 3d animatable avatars. Advances in Neural Information Processing Systems, 36, 2024. 3, 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.874, + 0.905, + 0.901 + ], + "angle": 0, + "content": "[24] Yangyi Huang, Hongwei Yi, Yuliang Xiu, Tingting Liao, Ji-axiang Tang, Deng Cai, and Justus Thies. Tech: Text-guided" + }, + { + "type": "list", + "bbox": [ + 0.517, + 0.093, + 0.905, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.945, + 0.521, + 0.957 + ], + "angle": 0, + "content": "26396" + } + ], + [ + { + "type": "ref_text", + "bbox": [ + 0.125, + 0.092, + 0.482, + 0.133 + ], + "angle": 0, + "content": "reconstruction of lifelike clothed humans. In 2024 International Conference on 3D Vision (3DV), pages 1531-1542. IEEE, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.134, + 0.483, + 0.189 + ], + "angle": 0, + "content": "[25] Bernhard Kerbl, Georgios Kopanas, Thomas Leimkuhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139-1, 2023. 3, 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.19, + 0.482, + 0.245 + ], + "angle": 0, + "content": "[26] Jeongho Kim, Gyojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on. arXiv preprint arXiv:2312.01725, 2023. 3, 4, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.247, + 0.482, + 0.273 + ], + "angle": 0, + "content": "[27] Diederik P Kingma. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.274, + 0.482, + 0.328 + ], + "angle": 0, + "content": "[28] Zorah Lahner, Daniel Cremers, and Tony Tung. Deepwrinkles: Accurate and realistic clothing modeling. In Proceedings of the European conference on computer vision (ECCV), pages 667-684, 2018. 1, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.33, + 0.482, + 0.398 + ], + "angle": 0, + "content": "[29] Sangyun Lee, Gyojung Gu, Sunghyun Park, Seunghwan Choi, and Jaegul Choo. High-resolution virtual try-on with misalignment and occlusion-handled conditions. In Proceedings of the European conference on computer vision (ECCV), 2022. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.4, + 0.482, + 0.455 + ], + "angle": 0, + "content": "[30] Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J Black. Smpl: A skinned multiperson linear model. In Seminal Graphics Papers: Pushing the Boundaries, Volume 2, pages 851-866. 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.457, + 0.482, + 0.524 + ], + "angle": 0, + "content": "[31] Yifang Men, Yiming Mao, Yuning Jiang, Wei-Ying Ma, and Zhouhui Lian. Controllable person image synthesis with attribute-decomposed gan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5084-5093, 2020. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.525, + 0.482, + 0.593 + ], + "angle": 0, + "content": "[32] Aymen Mir, Thiemo Alldieck, and Gerard Pons-Moll. Learning to transfer texture from clothing images to 3d humans. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7023-7034, 2020. 2, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.595, + 0.482, + 0.65 + ], + "angle": 0, + "content": "[33] Davide Morelli, Alberto Baldrati, Giuseppe Cartella, Marcelia Cornia, Marco Bertini, and Rita Cucchiara. Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on. arXiv preprint arXiv:2305.13501, 2023. 1, 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.651, + 0.482, + 0.719 + ], + "angle": 0, + "content": "[34] Maxime Oquab, Timothee Darcet, Theo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.72, + 0.483, + 0.803 + ], + "angle": 0, + "content": "[35] Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed AA Osman, Dimitrios Tzionas, and Michael J Black. Expressive body capture: 3d hands, face, and body from a single image. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10975-10985, 2019. 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.804, + 0.483, + 0.858 + ], + "angle": 0, + "content": "[36] Gerard Pons-Moll, Sergi Pujades, Sonny Hu, and Michael J Black. Clothcap: Seamless 4d clothing capture and retargeting. ACM Transactions on Graphics (ToG), 36(4):1-15, 2017. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.859, + 0.483, + 0.901 + ], + "angle": 0, + "content": "[37] Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Mildenhall. Dreamfusion: Text-to-3d using 2d diffusion. arXiv preprint arXiv:2209.14988, 2022. 2, 3" + }, + { + "type": "list", + "bbox": [ + 0.093, + 0.092, + 0.483, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.092, + 0.906, + 0.161 + ], + "angle": 0, + "content": "[38] Yurui Ren, Xiaoqing Fan, Ge Li, Shan Liu, and Thomas H Li. Neural texture extraction and distribution for controllable person image synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13535-13544, 2022. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.163, + 0.907, + 0.232 + ], + "angle": 0, + "content": "[39] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10684-10695, 2022. 1, 3, 4, 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.233, + 0.906, + 0.287 + ], + "angle": 0, + "content": "[40] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. arXiv preprint arXiv:2208.12242, 2022. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.288, + 0.907, + 0.357 + ], + "angle": 0, + "content": "[41] Igor Santesteban, Nils Thuerey, Miguel A Otaduy, and Dan Casas. Self-supervised collision handling via generative 3d garment models for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11763-11773, 2021. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.358, + 0.906, + 0.412 + ], + "angle": 0, + "content": "[42] Igor Santesteban, Miguel Otaduy, Nils Thuerey, and Dan Casas. Ulnef: Untangled layered neural fields for mix-and-match virtual try-on. Advances in Neural Information Processing Systems, 35:12110-12125, 2022. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.413, + 0.906, + 0.48 + ], + "angle": 0, + "content": "[43] Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International Conference on Machine Learning, pages 2256-2265. PMLR, 2015. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.482, + 0.906, + 0.524 + ], + "angle": 0, + "content": "[44] Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020.3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.525, + 0.906, + 0.566 + ], + "angle": 0, + "content": "[45] Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. Advances in Neural Information Processing Systems, 32, 2019. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.567, + 0.906, + 0.608 + ], + "angle": 0, + "content": "[46] Haoyu Wang, Zhilu Zhang, Donglin Di, Shiliang Zhang, and Wangmeng Zuo. Mv-vton: Multi-view virtual try-on with diffusion models. arXiv preprint arXiv:2404.17364, 2024. 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.609, + 0.906, + 0.676 + ], + "angle": 0, + "content": "[47] Junjie Wang, Jiemin Fang, Xiaopeng Zhang, Lingxi Xie, and Qi Tian. Gaussianeditor: Editing 3d gaussians delicately with text instructions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20902-20911, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.678, + 0.906, + 0.734 + ], + "angle": 0, + "content": "[48] Jing Wu, Jia-Wang Bian, Xinghui Li, Guangrun Wang, Ian Reid, Philip Torr, and Victor Adrian Prisacariu. Gaussctrl: multi-view consistent text-driven 3d gaussian splattering editing. arXiv preprint arXiv:2403.08733, 2024. 3, 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.735, + 0.906, + 0.816 + ], + "angle": 0, + "content": "[49] Jay Zhangjie Wu, Yixiao Ge, Xintao Wang, Stan Weixian Lei, Yuchao Gu, Yufei Shi, Wynne Hsu, Ying Shan, Xiaohu Qie, and Mike Zheng Shou. Tune-a-video: One-shot tuning of image diffusion models for text-to-video generation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 7623-7633, 2023. 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.818, + 0.906, + 0.871 + ], + "angle": 0, + "content": "[50] Zhenyu Xie, Haoye Dong, Yufei Gao, Zehua Ma, and Xiaodan Liang. Dreamvton: Customizing 3d virtual try-on with personalized diffusion models. arXiv preprint arXiv:2407.16511, 2024. 2, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.873, + 0.906, + 0.901 + ], + "angle": 0, + "content": "[51] Zhangyang Xiong, Chenghong Li, Kenkun Liu, Hongjie Liao, Jianqiao Hu, Junyi Zhu, Shuliang Ning, Lingteng Qiu," + }, + { + "type": "list", + "bbox": [ + 0.517, + 0.092, + 0.907, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "26397" + } + ], + [ + { + "type": "ref_text", + "bbox": [ + 0.125, + 0.092, + 0.483, + 0.161 + ], + "angle": 0, + "content": "Chongjie Wang, Shijie Wang, et al. Mvhumannet: A large-scale dataset of multi-view daily dressing human captures. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19801-19811, 2024. 2, 6, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.162, + 0.483, + 0.218 + ], + "angle": 0, + "content": "[52] Dejia Xu, Weili Nie, Chao Liu, Sifei Liu, Jan Kautz, Zhangyang Wang, and Arash Vahdat. Camco: Camera-controllable 3d-consistent image-to-video generation. arXiv preprint arXiv:2406.02509, 2024. 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.219, + 0.482, + 0.272 + ], + "angle": 0, + "content": "[53] Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen. Oot-diffusion: Outfitting fusion based latent diffusion for controllable virtual try-on. arXiv preprint arXiv:2403.01779, 2024. 3, 4, 8" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.274, + 0.483, + 0.328 + ], + "angle": 0, + "content": "[54] Han Yang, Xinrui Yu, and Ziwei Liu. Full-range virtual try-on with recurrent tri-level transform. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3460-3469, 2022. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.33, + 0.482, + 0.411 + ], + "angle": 0, + "content": "[55] Tao Yu, Zerong Zheng, Kaiwen Guo, Pengpeng Liu, Qionghai Dai, and Yebin Liu. Function4d: Real-time human volumetric capture from very sparse consumer rgbd sensors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5746-5756, 2021. 2, 6, 7, 8" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.413, + 0.482, + 0.468 + ], + "angle": 0, + "content": "[56] Jinsong Zhang, Kun Li, Yu-Kun Lai, and Jingyu Yang. Pise: Person image synthesis and editing with decoupled gan. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7982-7990, 2021. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.47, + 0.482, + 0.509 + ], + "angle": 0, + "content": "[57] Lvmin Zhang and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. arXiv preprint arXiv:2302.05543, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.511, + 0.482, + 0.566 + ], + "angle": 0, + "content": "[58] Yabo Zhang, Yuxiang Wei, Dongsheng Jiang, Xiaopeng Zhang, Wangmeng Zuo, and Qi Tian. Controlvideo: Training-free controllable text-to-video generation. arXiv preprint arXiv:2305.13077, 2023. 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.567, + 0.482, + 0.648 + ], + "angle": 0, + "content": "[59] Fuwei Zhao, Zhenyu Xie, Michael Kampffmeyer, Haoye Dong, Songfang Han, Tianxiang Zheng, Tao Zhang, and Xiaodan Liang. M3d-vton: A monocular-to-3d virtual try-on network. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 13239-13249, 2021. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.65, + 0.482, + 0.732 + ], + "angle": 0, + "content": "[60] Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Sahara, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two unets. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4606-4615, 2023. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.734, + 0.482, + 0.789 + ], + "angle": 0, + "content": "[61] Jingyu Zhuang, Chen Wang, Liang Lin, Lingjie Liu, and Guanbin Li. Dreameditor: Text-driven 3d scene editing with neural fields. In SIGGRAPH Asia 2023 Conference Papers, pages 1-10, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.79, + 0.482, + 0.843 + ], + "angle": 0, + "content": "[62] Jingyu Zhuang, Di Kang, Linchao Bao, Liang Lin, and Guanbin Li. Dagensm: Disentangled avatar generation with gs-enhanced mesh. arXiv preprint arXiv:2411.15205, 2024.3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.845, + 0.482, + 0.9 + ], + "angle": 0, + "content": "[63] Jingyu Zhuang, Di Kang, Yan-Pei Cao, Guanbin Li, Liang Lin, and Ying Shan. Tip-editor: An accurate 3d editor following both text-prompts and image-prompts. arXiv preprint arXiv:2401.14828, 2024. 3, 6, 7, 1" + }, + { + "type": "list", + "bbox": [ + 0.093, + 0.092, + 0.483, + 0.9 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.946, + 0.519, + 0.957 + ], + "angle": 0, + "content": "26398" + } + ] +] \ No newline at end of file diff --git a/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/4af7cfba-2431-4854-85f1-f3f911dcc811_origin.pdf b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/4af7cfba-2431-4854-85f1-f3f911dcc811_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d49cdd17c20fa7ca2b5aa2b805772d6fac1bc3e8 --- /dev/null +++ b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/4af7cfba-2431-4854-85f1-f3f911dcc811_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8beaca2d741a0c68bff5d3478c50c4d6e1879052f576cc81d91835be2382bf22 +size 2082731 diff --git a/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/full.md b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/full.md new file mode 100644 index 0000000000000000000000000000000000000000..2aa2d696bb0d4ac4337a0a0d95fbf2a1be9d350a --- /dev/null +++ b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/full.md @@ -0,0 +1,284 @@ +# VTON 360: High-Fidelity Virtual Try-On from Any Viewing Direction + +Zijian He1 Yuwei Ning1 Yipeng Qin2 Guangrun Wang1 Sibei Yang3 Liang Lin1,4,5 Guanbin Li1,4,5* +1Sun Yat-sen University +2Cardiff University +3ShanghaiTech University +4Guangdong Key Laboratory of Big Data Analysis and Processing +5Peng Cheng Laboratory +hezj39@mail2.sysu.edu.cn, yuwei.ning@hust.edu.cn, {qinyipengl991, wanggrun}@gmail.com, yangsb@shanghaiitech.edu.cn linliang@ieee.org, liguanbin@mail.sysu.edu.cn + +![](images/3e4ed0ca9fe9c0466bbb1e4f4019077b1eea386c4b063ea02f3897e6315ec9c9.jpg) +Figure 1. Results of VTON 360. Our VTON 360 enables high-fidelity 3D Virtual Try-On (VTON) by seamlessly adapting E-commerce garments onto a clothed 3D human model, supporting full $360^{\circ}$ view rendering. The highlighted bounding boxes (dashed line) demonstrate our method's ability to preserve intricate clothing details and patterns (e.g., collar accessories, horizontal line patterns, logos, texts, numbers) across diverse garment types. + +![](images/18ca224d94844f9dde5541e3e34a54b63839bcf1d69569779a98633de60a2a1f.jpg) + +![](images/a5889a307af41522f81b308115e028b9c3f02dce5c03771e67bcdd87075b042d.jpg) + +# Abstract + +Virtual Try-On (VTON) is a transformative technology in e-commerce and fashion design, enabling realistic digital visualization of clothing on individuals. In this work, we propose VTON 360, a novel 3D VTON method that addresses the open challenge of achieving high-fidelity VTON that supports any-view rendering. Specifically, we leverage the equivalence between a 3D model and its rendered multiview 2D images, and reformulate 3D VTON as an extension of 2D VTON that ensures 3D consistent results across multiple views. To achieve this, we extend 2D VTON models to include multi-view garments and clothing-agnostic human body images as input, and propose several novel techniques to enhance them, including: i) a pseudo-3D pose representation using normal maps derived from the SMPL-X 3D human model, ii) a multi-view spatial attention mechanism that models the correlations between features from different viewing angles, and iii) a multi-view CLIP embedding that enhances the garment CLIP features used in 2D VTON with camera information. Extensive experiments on large-scale real datasets and clothing images from e-commerce + +platforms demonstrate the effectiveness of our approach. +Project page: https://scnuhealthy.github.io/ +VTON360. + +# 1. Introduction + +Virtual Try-On (VTON) enables realistic digital visualization of clothing on individuals and has emerged as a transformative technology in e-commerce and fashion design. While significant research efforts have been made on 2D VTON solutions [8, 12, 19, 33, 39], these approaches are inherently limited in their representation of view-related features. To overcome this limitation and enable high-fidelity any-view rendering, 3D VTON methods were introduced. + +3D VTON requires accurate garment transfer onto a 3D human body while ensuring realistic garment fitting, texture preservation, and 3D consistency. The two primary aims of 3D VTON are i) achieving high-fidelity and ii) supporting any-view rendering. Leveraging the inherent capability of 3D models for any-view rendering, early 3D VTON methods [13, 15, 28] make clothing simulation on synthetic human bodies. Specifically, these methods utilized 3D scanners to capture clothing meshes, followed by the development of specialized dressing algorithms. Although effec + +tive, these methods rely on costly 3D scanning equipment and the physical presence of the human body/clothing (i.e., not fully virtual), restricting their practicality in real-world applications. As a byproduct, most early methods focused on developing geometrically correct dressing algorithms using standard templates of human body and clothing models. Addressing this limitation, researchers extended 3D VTON by introducing algorithms that reconstruct 3D clothing models from input images, enabling the use of image-based clothing inputs [3, 32, 41, 42]. However, since input clothing images (usually frontal) are inherently 2D and lack multi-view information, this approach struggles to reconstruct high-fidelity clothing models that can be rendered well from all viewing directions. + +To complement this missing information, DreamVTON [50] introduces a novel approach that leverages Text-to-Image (T2I) diffusion models to reconstruct both the human body and clothing from input images. Its key insight is that T2I models learned view-agnostic "concepts" of both bodies and garments during their training, and that the corresponding concepts for the input body and clothing images can be obtained using LoRA [21]. By utilizing Score Distillation Sampling (SDS) [37], DreamVTON can generate visual-pleasing 3D VTON results by ensuring consistency between renderings from arbitrary viewpoints and the concepts. Nonetheless, DreamVTON's high flexibility comes at the cost of low fidelity. This limitation stems from the fact that the concepts learned by T2I models are semantic in nature, thus lacking 3D geometric consistency and pixel-level accuracy with respect to the input body and clothing images. Recently, a concurrent work, namely GaussianVTON [6], partially addressed this limitation by formulating 3D VTON as a 3D scene editing task, where a given 3D human model is edited using multi-view images generated by 2D VTON methods. While it significantly enhances the fidelity of the human body, the fidelity and 3D consistency of clothing remain problematic, as there are no 2D VTON methods that can generate multi-view images with 3D consistency. Therefore, to the best of our knowledge, achieving high-fidelity 3D VTON that supports any-view rendering remains an open challenge. + +In this work, we address the above-mentioned challenge via proposing VTON 360, a novel 3D VTON method that achieves high-fidelity VTON from arbitrary viewing directions. Similar to GaussianVTON [6], our method edits a given 3D human model by inpainting the rendered images using a latent diffusion model. However, we set ourselves apart through our novel garment fidelity preservation strategy that can generate high-fidelity on-body garments in all viewing directions. Specifically, we first extend both the garment and clothing-agnostic human body inputs to typical 2D VTON models to leverage multi-view information, including paired front and back view garment images as well + +as a set of multi-view clothing-agnostic human body images sampled from random azimuth angles. Then, we propose several novel enhancements to bridge the gap between typical 2D VTON methods and our multi-view 3D consistency requirements: i) We propose a pseudo-3D pose representation using normal maps derived from the SMPL-X 3D human model, which captures fine-grained surface orientation details and provides more consistent geometry across views compared to the 2D pose representations (semantic segmentation maps) used in 2D VTON models. ii) We design a Multi-view Spatial Attention mechanism that models the correlations between features from different viewing angles, featuring a novel "correlation" matrix modeling the relationships among different input views. iii) We propose a multi-view CLIP embedding that enhances the garment CLIP embedding used in 2D VTON methods with camera information, thereby facilitating network learning of features relevant to a particular view. Together, these innovations enable our 2D VTON model to generate high-quality, multi-view and 3D-consistent virtual try-on results. Extensive experiments on Thuman2.0 [55] and MVHumanNet [51] datasets demonstrate that our method achieves high fidelity 3D VTON which supports any-view rendering. In addition, we show the effectiveness and generalizability of our methodology by testing it using garments from e-commerce platforms. Our conclusions include: + +- We propose a novel 3D Virtual Try-On (VTON) method, namely VTON 360, which achieves high-fidelity VTON from arbitrary viewing directions. +- Leveraging the equivalence between a 3D model and its rendered multi-view 2D images, we reformulate 3D VTON as an extension of 2D VTON that ensures consistent results across multiple views. Specifically, we introduce several novel techniques, including: (i) pseudo-3D pose representation; (ii) multi-view spatial attention; and (iii) multi-view CLIP embedding. These innovations enhance traditional 2D VTON models to generate multiview and 3D-consistent results. +- Extensive experimental results on two large real datasets as well as real clothing images from e-commerce platforms demonstrate the effectiveness of our approach. + +# 2. Related Work + +2D Virtual Try-On. 2D Virtual Try-On (VTON) aims to transfer a desired garment to the corresponding region of a target human image while preserving the human pose and identity. Early methods [2, 8, 10, 11, 16, 18, 29, 31, 38, 54, 56] use Generative Adversarial Networks (GANs) to deform the garments to match the target body shape, which a critical step for achieving realistic VTON. However, accurately adapting to diverse real-world conditions remains a significant challenge. Addressing this issue, recent methods [12, 19, 33, 60] reframe 2D VTON as a conditioned in- + +painting task, leveraging the strong priors provided by diffusion models [20, 43, 45] to achieve promising results. This strategy is further improved by [9, 26, 53], which introduce a ReferenceNet to extract hierarchical garment features and apply attention mechanisms to condition the Main UNet. + +3D Virutal Try-On. For 3D Virtual Try-On (VTON), traditional methods [4, 13, 15, 28, 36] rely on 3D scanning or cloth simulation to generate highly precise body and garment geometry. These methods were then extended by learning-based methods [3, 32] that employ differentiable rendering to dress the SMPL [30] model with a desired garment mesh. Despite their effectiveness, such methods rely on costly 3D scanning and the physical presence of human body/clothing, limiting their application in the real world. Addressing this limitation, M3D-VTON [59] proposes a depth-based 3D VTON framework to reconstruct 3D clothed human models from 2D human and garment images, but the results often suffer from explicit warping artifacts. To improve 3D VTON results, recent methods [23, 24, 50, 62] resort to text-to-image (T2I) diffusion models and employ the Score Distillation Sampling (SDS) loss [40] to ensure consistency among different viewing directions. Specifically, TeCH [24] adapts the generative priors of T2I diffusion model to the specific person and clothes by training descriptive text prompts with DreamBooth [40]. DreamWaltz [23] leverages Pose ControlNet [57] to attain clothed human body models. DreamVTON [50] introduces a multi-concept LoRA [21] to personalize the T2I diffusion model, and uses a template-based optimization mechanism that combines with SDS loss to better preserve patterns on the garment. Although effective, these methods often produce results lacking in fidelity, as the concepts learned by T2I models are semantic rather than at the pixel level. Concurrent to our work, GaussianVTON [6] proposes an alternative approach by combining Gaussian Splatting [25] with pre-trained 2D VTON models and formulate it as an editing task. However, since there are no 2D VTON methods that can generate multi-view images with 3D consistency, the fidelity and 3D consistency of the clothing generated remain problematic. + +Radiance Field-based 3D Human or Scene Editing. Recently, radiance field-based editing has attracted significant interest due to its efficient differentiable rendering capabilities, sparking substantial advancements in text-driven 3D editing. For example, InstructN2N [17] employ an image-based diffusion model InstructP2P [5] to modify the rendered image by the user's text description with a variant of the score distillation sampling (SDS) [37] loss. GaussianEditor [7] applies Gaussian Splatting [25] as 3D representation instead of NeRF, adopting Gaussian semantic tracking to track target Gaussian values, significantly improving editing speed and controllability. To enable accu + +rate location and appearance control, subsequent works [47, 61] specify the editing region using the attention score or with a segmentation model. TIP-Editor [63] proposes a content personalization step dedicated to the reference image based on LoRA, achieving the editing following a hybrid text-image prompt. GaussCtrl [48] leverage depth conditions and attention-based latent code alignment to achieve 3D-aware multi-view consistent editing instead of iteratively editing single views using SDS loss. However, these works primarily focus on global appearance modifications within a text-driven pipeline, while our approach emphasizes preserving fine textural details from different viewing directions throughout the editing process. + +# 3. Preliminary + +Latent Diffusion Model. Latent Diffusion Model [39] is a variant of diffusion models that performs denoising within the latent space of a Variational Autoencoder (VAE) [27]. During training, given a fixed encoder $\mathcal{E}$ , an input image $x$ is transformed into its latent representation $z_0 = \mathcal{E}(x)$ . A conditional diffusion model $\hat{\epsilon}_{\theta}$ , typically implemented with a UNet architecture, is then trained using a weighted denoising score matching objective: + +$$ +\mathcal {L} _ {L D M} = \mathbb {E} _ {\mathbf {z}, \mathbf {c}, \boldsymbol {\epsilon}, t} [ \boldsymbol {\epsilon} - \| \hat {\epsilon} _ {\theta} (\mathbf {z} _ {\mathbf {t}}; \mathbf {c}, t) \| _ {2} ^ {2} ] \tag {1} +$$ + +where $\mathbf{z}_t\coloneqq \alpha_t\mathbf{x} + \sigma_t\boldsymbol {\epsilon}$ denotes the forward diffusion process at timestep $t$ . $\alpha_{t},\sigma_{t}$ are time-dependent functions defined by the diffusion model formulation; c denotes the conditional input and $\epsilon \sim \mathcal{N}(\mathbf{0},\mathbf{1})$ is Gaussian noise. During inference, data samples are generated by initiating from Gaussian noise $\mathbf{z}_T\sim \mathcal{N}(\mathbf{0},\mathbf{1})$ and iteratively refining it using a DDIM [44] sampler. + +# 4. Method + +Our method leverages the equivalence between a 3D model and its rendered multi-view 2D images to achieve high-fidelity, any-view 3D VTON. Specifically, as Fig. 2 shows, given an input 3D human model and a garment image, our method 1) renders the 3D model into multi-view 2D images and 2) formulates 3D VTON as a consistent, unified 2D VTON process across these rendered views; 3) By reconstructing the edited images into a 3D model using existing 3D reconstruction methods, we ensure visual coherence and precise garment alignment from any viewing angle. Among them, the second step is crucial as existing 2D VTON methods [9, 26, 53] lack 3D knowledge, preventing them from generating multi-view images with 3D consistency. + +To address this challenge, we propose several novel techniques (Sec. 4.2) that equip a typical 2D VTON network (Sec. 4.1), which is built on a latent diffusion model [39], with the capability to generate 3D-consistent results. We use Gaussian Splatting [25] as our 3D representation. + +![](images/50456d17e5ccbb76f3e583fe2b0ebac3fa363e2d03f09e2a13d08f82ca40471e.jpg) +Figure 2. Overview of VTON 360. Given an input 3D human model $\mathbf{G}_{\mathrm{src}}$ and a pair of garment images $(g_{f}, g_{b})$ , our method 1) renders $\mathbf{G}_{\mathrm{src}}$ into multi-view 2D images (left) and 2) edits the rendered multi-view 2D images (middle); 3) reconstructs the edited images into a 3D model $\mathbf{G}_{\mathrm{VTON}}$ (right). In the crucial step 2), we propose three novel techniques to equip a typical 2D VTON network with the capability to generate 3D-consistent results: 1) Pseudo-3D Pose Input, 2) Multi-view Spatial Attention, and 3) Multi-view CLIP Embedding. + +![](images/c3dd68bc18957e8ae4d2aa15c28ee2931189de321d4ce4c838cc8d0a92c0acbb.jpg) +Figure 3. DensePose (2D) vs. SMPL-X normal map (pseudo-3D) representations. DensePose applies uniform labels per body part, lacking 3D consistency across views and causing artifacts and temporal inconsistencies (highlighted with red boxes). In contrast, SMPL-X normal maps capture fine surface details, ensuring geometric coherence and stable, realistic shading across views. + +# 4.1. Recap of 2D VTON Framework + +Following previous works [12, 26, 53], we formulate 2D VTON as an exemplar-based inpainting problem, aiming to fill an input clothing-agnostic image $\mathbf{A}$ with a given garment image $g$ , where $\mathbf{A}$ is obtained following the method used in [53]. As illustrated in Fig. 2 (middle), the network architecture is based on the latent diffusion model [39] with + +an encoder $\mathcal{E}$ and comprises two components: + +- A GarmentNet [9, 53] that extracts features from $\mathcal{E}(g)$ . +- A Main UNet that inpaints $\mathbf{A}$ according to i) detailed garment features extracted by the GarmentNet; ii) the 2D pose of $\mathbf{A}$ represented by semantic labels using Dense-Pose [14]; iii) CLIP embeddings of input garment $g$ . Among them, i) and ii) together with noise are input to the self-attention layers, while iii) is input to the cross-attention layers of the Main UNet. + +Both the GarmentNet and the Main UNet share the same network architecture. + +# 4.2. Multi-view 2D VTON with 3D Consistency + +To enable the aforementioned 2D VTON model to generate multi-view and 3D-consistent results, we propose the following novel enhancements to its design: + +Multi-view Inputs. We extend both inputs to the model: + +- Multi-view Garment Inputs: We extend the input garment representation from a single image $g$ to paired front and back view images $g_{f}$ , $g_{b}$ , providing comprehensive garment information across all viewing angles. Accordingly, we use the encoder $\mathcal{E}$ to encode $g_{f}, g_{b}$ into their latent representations $\mathcal{E}(g_{f}), \mathcal{E}(g_{b})$ and feed them into GarmentNet to obtain their multi-layer features $F_{f}^{l}$ and $F_{b}^{l}$ at layer $l$ , respectively. +- Multi-view Clothing-agnostic Image Inputs: Based on + +the equivalence between a 3D human model and its rendered multi-view 2D images, we extend the input human body representation from a single, clothing-agnostic image, $\mathbf{A}$ , to a set of $m$ multi-view images, denoted as $\mathbf{A}_1, \mathbf{A}_2, \dots, \mathbf{A}_m$ . These images are sampled from random azimuth angles, allowing the 2D VTON model to access comprehensive, multi-view information from the input 3D human model. + +Pseudo-3D Pose Input. As shown in Fig. 3, the 2D Dense-Pose representations [14] commonly used in state-of-the-art 2D VTON methods [9, 26] assign a uniform semantic label to all pixels within each body part (e.g., thigh), inherently lack 3D geometric consistency across multiple views, and often introduce artifacts and temporal inconsistencies. To address these limitations, we propose a novel pseudo-3D pose representation: the normal maps $\mathbf{N}$ derived from the SMPL-X [35] model of the input body. These normal maps capture fine-grained surface orientation details, providing a more consistent representation across views by preserving geometric structure in the 3D space. Furthermore, they enable smoother, temporally stable transitions and realistic shading effects across multi-view scenarios. In practice, we employ a lightweight PoseEncoder $\mathcal{E}'$ [22] and feed $\mathcal{E}'(\mathbf{N})$ into the Main UNet. We obtain the SMPL-X model from the multi-view images of the input body using EasyMoCap [1]. + +Accordingly, we concatenate three components as the enhanced input to the Main UNet: i) a noise latent $z_{t}$ ; ii) the encoded pseudo-3D poses $\mathcal{E}'(\mathbf{N_1}),\mathcal{E}'(\mathbf{N_2}),\dots,\mathcal{E}'(\mathbf{N_m})$ ; and iii) the encoded multi-view clothing-agnostic images $\mathcal{E}(\mathbf{A_1}),\mathcal{E}(\mathbf{A_2}),\dots,\mathcal{E}(\mathbf{A_m})$ . Let $F_{1}^{l},F_{2}^{l},\dots,F_{m}^{l}$ be the feature representations at layer $l$ of the Main UNet, and recall the garment features $F_{f}^{l}$ and $F_{b}^{l}$ defined above, we enhance the self-attention layers of the Main UNet as: + +Multi-view Spatial Attention. To cope with the aforementioned multi-view input features and ensure their consistency, we draw insights from the temporal attention layer commonly used in video generation and editing [49, 58] and extend it to our multi-view spatial attention layer, denoted as MVAttention. The key distinction of our MVAttention is that its input multi-view features $F_{1}^{l}, F_{2}^{l}, \ldots, F_{m}^{l}$ are from images captured from non-uniform spatial intervals, with the viewing angles varying randomly. Consequently, features from similar views exhibit a higher correlation, while those from distinct views are largely independent. To model this relationship, we construct a "correlation" matrix $C$ based on the angular disparity obtained from camera rotation matrices of the input multi-view images, and define our MVAttention as follows: + +$$ +\begin{array}{l} \mathbf {F} ^ {\mathbf {l}} = \left[ F _ {1} ^ {l} \oplus F _ {2} ^ {l} \dots \oplus F _ {m} ^ {l} \right], \hat {\mathbf {F}} ^ {\mathbf {l}} = \left[ \mathbf {F} ^ {\mathbf {l}} \oplus F _ {f} ^ {l} \oplus F _ {b} ^ {l} \right] \\ Q = W ^ {Q} \mathbf {F} ^ {1}, K = W ^ {K} \hat {\mathbf {F}} ^ {1}, V = W ^ {V} \hat {\mathbf {F}} ^ {1} \tag {2} \\ A _ {i} = \operatorname {s o f t m a x} \left(\frac {Q _ {i} \times \left(C _ {i} \cdot K ^ {T}\right)}{\sqrt {d}}\right), \mathbf {H} _ {\mathbf {i}} ^ {\mathbf {l}} = A _ {i} \times V _ {i} \\ \end{array} +$$ + +where $i\in \{1,2,\dots,m\}$ denotes $i$ -th view; the Query $Q$ comes directly from $\mathbf{F}^1$ and the concatenation of $[\mathbf{F}^1,F_f^l,F_b^l ]$ serves as the key $K$ and the value $V$ . $\oplus$ indicates matrix concatenation along the token axis; $d$ denotes the dimension; $W^{Q},W^{K},W^{V}$ represent the linear transformation matrices; we omitted the $l$ of the attention matrices and parameters for simplicity; $C\in \mathbb{R}^{m\times m}$ $C_i$ represents $i$ -th row in $C$ , and its "correlation" value between $i$ -th and $j$ -th features is $C_{ij}$ .. + +$$ +C _ {i j} = \left(\left(\operatorname {t r a c e} \left(R _ {i} ^ {T} R _ {j}\right) - 1\right) / 2 + 1\right) / 2 \tag {3} +$$ + +where $R_{i}$ and $R_{j}$ are the extrinsic rotation matrices of the corresponding camera views, $(\mathrm{trace}(R_i^T R_j) - 1) / 2$ is the cosine value of the angle between these camera views. + +Multi-view CLIP Embedding. Camera viewpoints can serve as an effective condition signal to enhance 3D consistency in video content generation [52]. Building on this insight, we incorporate camera condition within our try-on network by encoding camera parameters as an additional token, enabling the generation of more consistent multi-view images. Specifically, we define a world coordinate system in which the camera faces the subject directly. For each input image (view) $\mathbf{A}_{\mathrm{i}}$ , $1 \leq i \leq m$ , we extract the rotation matrix from the camera's corresponding extrinsic matrix. This rotation matrix is then reshaped into a 9-dimensional tensor $\mathbf{r}_{\mathrm{i}}$ , which undergoes positional encoding to effectively integrate the camera parameters into the feature representation $F_{i}^{c}$ . + +$$ +\begin{array}{l} F _ {i} ^ {c} = \left(\sin \left(2 ^ {0} \pi \mathbf {r} _ {\mathbf {i}}\right), \cos \left(2 ^ {0} \pi \mathbf {r} _ {\mathbf {i}}\right), \dots , \right. \tag {4} \\ \sin (2 ^ {L - 1} \pi \mathbf {r _ {i}}), \cos (2 ^ {L - 1} \pi \mathbf {r _ {i}})) \\ \end{array} +$$ + +where $L$ is the length of positional embedding. We then project $F_{i}^{c}$ to match the dimensionality of the garment CLIP image embedding $F^{g}$ via an MLP and concatenate them along the token axis to form $Y_{i}$ . This combined representation, $Y_{i}$ , is subsequently used in the key $K_{\mathrm{x}}$ and value $V_{\mathrm{x}}$ of the cross-attention layers of the Main UNet: + +$$ +Y _ {i} = F ^ {g} \oplus \operatorname {M L P} \left(F _ {i} ^ {c}\right) +$$ + +$$ +Q _ {\mathrm {x}} = W _ {\mathrm {x}} ^ {Q} \mathbf {H} _ {\mathrm {i}} ^ {1}, K _ {\mathrm {x}} = W _ {\mathrm {x}} ^ {K} Y _ {i}, V _ {\mathrm {x}} = W _ {\mathrm {x}} ^ {V} Y _ {i} \tag {5} +$$ + +$$ +F _ {i} ^ {(l + 1)} = \mathrm {s o f t m a x} (\frac {Q _ {\mathrm {x}} K _ {\mathrm {x}} ^ {T}}{\sqrt {d _ {\mathrm {x}}}}) V _ {\mathrm {x}}, +$$ + +where $\mathbf{H}^1$ is the output of the MVAttention of the $l$ -th layer; we omitted the $l$ of the cross attention matrices and parameters for simplicity. + +Training. Our enhanced multi-view 2D VTON network can be trained by minimizing the following latent diffusion model loss function: + +$$ +\mathcal {L} _ {\mathrm {l d m}} = \mathbb {E} _ {z _ {t}, \eta , \psi , \epsilon , t} \left[ \| \epsilon - \epsilon_ {\theta} (z _ {t}, t, \eta , \psi , \zeta)) \| _ {2} ^ {2} \right], \tag {6} +$$ + +![](images/66efb9bdfdf378e1471d23c74712a4482299aa59ce3a5ebfeba941baa74bac58.jpg) +Figure 4. Illustration of the proposed Multi-view Spatial Attention. Query (Q): multi-view features $\mathbf{F}^1$ ; Key (K) and Value (V): concatenation of $\mathbf{F}^1$ and garment features $F_{f}^{l}$ and $F_{b}^{l}$ . The attention score between viewpoints $i$ and $j$ is modulated by a weight $C_{ij}$ , determined by the cosine of the angle between them. + +where $\eta = [\mathcal{E}(g_f); \mathcal{E}(g_b); \mathcal{E}(\mathbf{N_i})_{i=1}^m]$ represents the input latent garment images and latent normal maps; $\zeta = [\mathcal{E}'(\mathbf{A_i})_{i=1}^m]$ denotes the input latent clothing-agnostic images; $\psi = \mathbf{Y}$ is the proposed multi-view CLIP embedding. + +# 5. Experiments + +# 5.1. Experimental Setup + +Datasets. We conduct experiments on two public datasets: Thuman2.0 [55] and MVHumanNet [51]. Thuman2.0 comprises 526 reconstructed clothed human scans, from which we render multi-view input images. Of these samples, 426 are used for training, while the remaining 100 are set aside for testing. To further evaluate the effectiveness and robustness of our method, we also perform experiments on MVHumanNet, a large-scale dataset of multi-view human images that encompasses a diverse range of subjects, daily outfits and motion sequences. The images in MVHumanNet are captured using a multi-view system with either 48 or 24 cameras. We use 4,990 subjects from this dataset, allocating 4,790 to training and 200 for tests. For each subject, we randomly select two frames of multi-view images from its entire motion sequence. While MVHumanNet provides multi-view images directly for editing and reconstruction, we render uniformly distributed views around each human subject in Thuman2.0 to ensure consistent input. + +Baselines. We primarily compare our method with three existing methods: DreamWaltz [23], GaussCtrl [48], and TIP-Editor [63]. DreamWaltz is a method designed for directly generating 3D human bodies based on textual descriptions, while GaussCtrl and TIP-Editor are two radiance-based editing methods. GaussCtrl is based on Stable Diffusion, using a description-like prompt to edit the scene. TIP-Editor accepts both text and image prompts. We configure it by + +specifying the human body as the editing region and the desired garment as the image prompt. We use ChatGPT to generate the text prompts corresponding to the clothing images. + +Evaluation Metrics. For quantitative evaluation, we assess garment-to-person alignment between the edited person and the reference image. Following [63], we calculate the average DINO similarity [34] between the reference image and the rendered multi-view images of the edited 3D scene. Additionally, to evaluate multi-view consistency, we compute the CLIP Directional Consistency Score as outlined in [17]. Given the large scale of experiments (repeated 3DGS reconstruction), we selected a subset of examples from the dataset for metric evaluation. Specifically, from the test sets of Thuman and MVHumanNet, we randomly sampled 10 human scans each, performing virtual try-on with 6 randomly chosen garments per human scan. + +We further conducted a user study involving 50 participants who rated the results of our method and three baseline methods based on two criteria: overall "Quality" and "Alignment" with the reference image. Each evaluation consisted of two questions: (1) Which method produces the highest quality of the edited 3D human? and (2) Which method achieves the most consistent alignment with the target clothing? Participants viewed the VTON results as rotating randomized video sequences. + +Implementation Details. During pre-processing, we crop the multi-view images to the bounding box around the person and resize them to a resolution of $768 \times 576$ . The front view and the back view of garment images are obtained from the corresponding clothed human images. After editing, we pad the images back to their original size. The data processing pipeline is the same for both Thuman2.0 and MVHumanNet datasets. + +The Main UNet and the GarmentNet are initialized by the Stable Diffusion V1.5 [39]. The training process is divided into two stages. In the first stage, each view is trained independently, during which we establish the feature extraction capabilities of both the PoseEncoder and GarmentNet, as well as the generative capability of the Main UNet. The second stage involves multi-view training, where we randomly select $M$ views for each human subject. This stage is focused on training the proposed MVAttention module to enhance the network's ability to maintain consistency across views. Due to memory constraints, we set $M = 8$ for the training phase. During the testing phase, we uniformly sampled 32 views from a 360-degree rotation around the subject. The editing of these 32 views is conducted in two batches, with each batch processing $M = 16$ views. + +# 5.2. Comparisons with State-of-the-Art Methods + +Qualitative Evaluation. Fig. 5 shows visual comparisons between our method and the baselines. DreamWaltz [23] + +
MethodThuman2.0 [55]MVHumanNet [51]
CLIPcons↑DINOsim↑VotequalityVotealignCLIPcons↑DINOsim↑VotequalityVotealign
DreamWaltz [23]0.8870.5560.46%1.54%0.9350.4950.46%0.46%
TIP-Editor [63]0.9390.5690.92%0.62%0.9480.5122.15%1.38%
GaussCtrl [48]0.9310.5771.08%1.38%0.9380.5211.69%1.23%
Ours0.9230.63397.54%96.46%0.9330.62395.69%96.92%
+ +Table 1. Quantitative comparisons. ${\mathrm{{CLIP}}}_{\text{cons }}$ denotes the CLIP Direction Consistency Score. ${\mathrm{{DINO}}}_{\text{sim }}$ is the DINO similarity. + +![](images/6b71985790ddfa2ccbcfe297ed7bf36c95d426c4edc7ae3f9d723d3ce1a7d5c3.jpg) +Figure 5. Qualitative comparison. The first two rows show the results on Thuman2.0 dataset while the last two rows show the results on MVHumaNet dataset. Our method achieves good texture preservation (highlighted by the blue boxes), while three baseline methods mostly fail. + +regenerates 3D clothed humans from text prompts but struggles to accurately retain both body and clothing characteristics. GaussCtrl [48], lacking support for image prompts, fails to maintain detailed clothing textures. While TipEditor [63] leverages Lora [21] for personalization, it encounters difficulties in consistently mapping clothing inputs from two views into the 3D human because the personalized concept are semantic in 2D space. In contrast, our method effectively preserves intricate clothing details, such as text, stripes, and logos. + +Quantitative Evaluation. Tab. 1 shows the results for the CLIP Directional Consistency Score and DINO similarity on Thuman2.0 and MVHumanNet datasets. Our approach surpasses other methods on $\mathrm{DINO}_{sim}$ , clearly illustrating + +the superiority of our method in terms of garment texture preservation. While our results on $\mathrm{CLIP}_{\text {cons }}$ are comparable to those of other methods, it is important to note that these methods incorporate the SDS loss, which to some extent smoothes the representation of humans in 3D space. Additionally, the "flatter" textures of other methods could also result in artificially higher consistency scores. Furthermore, user studies have shown that our method significantly exceeds baselines in terms of edited 3D human quality and the alignment of clothing details. + +# 5.3. Visual Results using E-commerce Garment + +Fig. 6 showcases VTON results using garments from the MVG dataset [46], whose images are from e-commerce + +![](images/7737d81fbc34085b8a538160848be10370bc9d5255819ab87e19ccf3b3907f3f.jpg) +Figure 6. Generalization to e-commerce garments (the MVG dataset). Our method, trained on the THuman2.0 dataset, shows strong generalizability when applied to e-commerce garments. For clarity in visualization, we display garment images on human models; however, in the actual VTON process, the garments are segmented from the models using parse maps. + +platforms like YOOX NET-A-PORTER, Taobao, and TikTok*, and a model trained on the Thuman2.0 dataset [55]. The results demonstrate that our method effectively preserves intricate garment details and textures. For instance, it accurately maintains the stripe patterns in the first row, the cute tie in the second row, and the buttons in the third row, highlighting the robustness of our approach in handling diverse and realistic clothing items. + +# 5.4. Ablation Study + +We conduct an ablation study on Thuman2.0 dataset in Tab. 2 and Fig. 7 to evaluate the impact of our three proposed modules in enhancing a typical 2D VTON network with 3D-consistent generation capabilities. Starting with the 2D VTON baseline [53] using DensePose, we progressively replace DensePose with our pseudo-3D pose, incorporate multi-view CLIP embeddings, and ultimately integrate MVAttention in the final configuration. Results in Tab. 2 indicate that each module contributes to metric improvements. Fig. 7 visualizes an example of multi-view image editing. The incorporation of pseudo-3D pose substantially improves limb generation compared to the 2D VTON baseline. Comparing rows 4 and 5, prior to the integration of multi-view CLIP embedding, the model captures limited spatial information, resulting in detail loss at specific angles (columns 3, 4, and 6). Finally, the proposed MVAttention achieves a more coherent generation across views. + +![](images/5de364b5b01530170b9e9a96febaa852e03b779b915dc80ff95da76bb7a36c53.jpg) +Figure 7. Visualization of the impact of the three proposed techniques on multi-view consistent editing. The red boxes highlight the artifacts. Starting from the 2D VTON baseline, the pseudo-3D pose improves limb generation quality, multi-view CLIP embedding enhances detail across different viewing directions, and finally, MVAttention further strengthens consistency in the generated images. + +
MethodsCLIPcons↑DINOsim↑
2D-VTON0.8920.609
+ Pseudo-3D Pose0.9100.626
+ Multi-view CLIP Embedding0.9130.631
+ Multi-view Spatial Attention0.9230.633
+ +Table 2. Ablation studies. We ablate the impact of the three proposed techniques on Thuman2.0 dataset. + +# 6. Conclusions + +In this work, we proposed VTON 360, a novel 3D Virtual Try-On (VTON) method that achieves high-fidelity VTON with the ability to render clothing from arbitrary viewing directions. Our method features a novel formulation of 3D VTON as an extension of 2D VTON that ensures 3D consistent results across multiple views. To bridge the gap between 2D VTON models and 3D consistency requirements, we introduce several key innovations, including multi-view inputs, pseudo-3D pose representation, multi-view spatial attention, and multi-view CLIP embedding. Extensive experiments demonstrate the effectiveness of our approach, significantly outperforming prior 3D VTON techniques in both fidelity and any-view rendering. + +# Acknowledgement + +This work is supported in part by the National Key R&D Program of China under Grant No.2024YFB3908503, in part by the National Natural Science Foundation of China under Grant NO. 62322608 and in part by the CCF-Kuaishou Large Model Explorer Fund (NO. CCF-KuaiShou 2024007). + +# References + +[1] Easymocap - make human motion capture easier. Github, 2021.5 +[2] Shuai Bai, Huiling Zhou, Zhikang Li, Chang Zhou, and Hongxia Yang. Single stage virtual try-on via deformable attention flows. In European Conference on Computer Vision, pages 409-425. Springer, 2022. 2 +[3] Bharat Lal Bhatnagar, Garvita Tiwari, Christian Theobalt, and Gerard Pons-Moll. Multi-garment net: Learning to dress 3d people from images. In Proceedings of the IEEE/CVF international conference on computer vision, pages 5420-5430, 2019. 2, 3 +[4] Robert Bridson, Ronald Fedkiw, and John Anderson. Robust treatment of collisions, contact and friction for cloth animation. In Proceedings of the 29th annual conference on Computer graphics and interactive techniques, pages 594-603, 2002. 3 +[5] Tim Brooks, Aleksander Holynski, and Alexei A Efros. Instructpix2pix: Learning to follow image editing instructions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18392-18402, 2023. 3 +[6] Haodong Chen, Yongle Huang, Haojian Huang, Xiangsheng Ge, and Dian Shao. Gaussianvton: 3d human virtual try-on via multi-stage gaussian splatting editing with image prompting. arXiv preprint arXiv:2405.07472, 2024. 2, 3, 1 +[7] Yiwen Chen, Zilong Chen, Chi Zhang, Feng Wang, Xiaofeng Yang, Yikai Wang, Zhongang Cai, Lei Yang, Huaping Liu, and Guosheng Lin. Gaussianeditor: Swift and controllable 3d editing with gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21476-21485, 2024. 3 +[8] Seunghwan Choi, Sunghyun Park, Minsoo Lee, and Jaegul Choo. Viton-hd: High-resolution virtual try-on via misalignment-aware normalization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14131-14140, 2021. 1, 2 +[9] Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for virtual try-on. arXiv preprint arXiv:2403.05139, 2024. 3, 4, 5 +[10] Xin Dong, Fuwei Zhao, Zhenyu Xie, Xijin Zhang, Daniel K Du, Min Zheng, Xiang Long, Xiaodan Liang, and Jianchao Yang. Dressing in the wild by watching dance videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3480-3489, 2022. 2 + +[11] Yuying Ge, Yibing Song, Ruimao Zhang, Chongjian Ge, Wei Liu, and Ping Luo. Parser-free virtual try-on via distilling appearance flows. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8485-8493, 2021. 2 +[12] Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang. Taming the power of diffusion models for high-quality virtual try-on with appearance flow. arXiv preprint arXiv:2308.06101, 2023. 1, 2, 4 +[13] Peng Guan, Loretta Reiss, David A Hirshberg, Alexander Weiss, and Michael J Black. Drape: Dressing any person. ACM Transactions on Graphics (ToG), 31(4):1-10, 2012. 1, 3 +[14] Riza Alp Güler, Natalia Neverova, and Iasonas Kokkinos. Densepose: Dense human pose estimation in the wild. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7297-7306, 2018. 4, 5 +[15] Fabian Hahn, Bernhard Thomaszewski, Stelian Coros, Robert W Sumner, Forrester Cole, Mark Meyer, Tony DeRose, and Markus Gross. Subspace clothing simulation using adaptive bases. ACM Transactions on Graphics (TOG), 33(4):1-9, 2014. 1, 3 +[16] Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S Davis. Viton: An image-based virtual try-on network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7543-7552, 2018. 2 +[17] Ayaan Haque, Matthew Tancik, Alexei A Efros, Aleksander Holynski, and Angjoo Kanazawa. Instruct-nerf2nerf: Editing 3d scenes with instructions. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 19740–19750, 2023. 3, 6, 1 +[18] Sen He, Yi-Zhe Song, and Tao Xiang. Style-based global appearance flow for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3470-3479, 2022. 2 +[19] Zijian He, Peixin Chen, Guangrun Wang, Guanbin Li, Philip HS Torr, and Liang Lin. Wildvidfit: Video virtual try-on in the wild via image-based controlled diffusion models. In European Conference on Computer Vision, pages 123-139. Springer, 2024. 1, 2 +[20] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in Neural Information Processing Systems, 33:6840-6851, 2020. 3 +[21] Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021. 2, 3, 7 +[22] Li Hu, Xin Gao, Peng Zhang, Ke Sun, Bang Zhang, and Liefeng Bo. Animate anyone: Consistent and controllable image-to-video synthesis for character animation. arXiv preprint arXiv:2311.17117, 2023. 5 +[23] Yukun Huang, Jianan Wang, Ailing Zeng, He Cao, Xianbiao Qi, Yukai Shi, Zheng-Jun Zha, and Lei Zhang. Dreamwaltz: Make a scene with complex 3d animatable avatars. Advances in Neural Information Processing Systems, 36, 2024. 3, 6, 7 +[24] Yangyi Huang, Hongwei Yi, Yuliang Xiu, Tingting Liao, Ji-axiang Tang, Deng Cai, and Justus Thies. Tech: Text-guided + +reconstruction of lifelike clothed humans. In 2024 International Conference on 3D Vision (3DV), pages 1531-1542. IEEE, 2024. 3 +[25] Bernhard Kerbl, Georgios Kopanas, Thomas Leimkuhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139-1, 2023. 3, 1 +[26] Jeongho Kim, Gyojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on. arXiv preprint arXiv:2312.01725, 2023. 3, 4, 5 +[27] Diederik P Kingma. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013. 3 +[28] Zorah Lahner, Daniel Cremers, and Tony Tung. Deepwrinkles: Accurate and realistic clothing modeling. In Proceedings of the European conference on computer vision (ECCV), pages 667-684, 2018. 1, 3 +[29] Sangyun Lee, Gyojung Gu, Sunghyun Park, Seunghwan Choi, and Jaegul Choo. High-resolution virtual try-on with misalignment and occlusion-handled conditions. In Proceedings of the European conference on computer vision (ECCV), 2022. 2 +[30] Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J Black. Smpl: A skinned multiperson linear model. In Seminal Graphics Papers: Pushing the Boundaries, Volume 2, pages 851-866. 2023. 3 +[31] Yifang Men, Yiming Mao, Yuning Jiang, Wei-Ying Ma, and Zhouhui Lian. Controllable person image synthesis with attribute-decomposed gan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5084-5093, 2020. 2 +[32] Aymen Mir, Thiemo Alldieck, and Gerard Pons-Moll. Learning to transfer texture from clothing images to 3d humans. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7023-7034, 2020. 2, 3 +[33] Davide Morelli, Alberto Baldrati, Giuseppe Cartella, Marcelia Cornia, Marco Bertini, and Rita Cucchiara. Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on. arXiv preprint arXiv:2305.13501, 2023. 1, 2 +[34] Maxime Oquab, Timothee Darcet, Theo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 6 +[35] Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed AA Osman, Dimitrios Tzionas, and Michael J Black. Expressive body capture: 3d hands, face, and body from a single image. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10975-10985, 2019. 5 +[36] Gerard Pons-Moll, Sergi Pujades, Sonny Hu, and Michael J Black. Clothcap: Seamless 4d clothing capture and retargeting. ACM Transactions on Graphics (ToG), 36(4):1-15, 2017. 3 +[37] Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Mildenhall. Dreamfusion: Text-to-3d using 2d diffusion. arXiv preprint arXiv:2209.14988, 2022. 2, 3 + +[38] Yurui Ren, Xiaoqing Fan, Ge Li, Shan Liu, and Thomas H Li. Neural texture extraction and distribution for controllable person image synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13535-13544, 2022. 2 +[39] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10684-10695, 2022. 1, 3, 4, 6 +[40] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. arXiv preprint arXiv:2208.12242, 2022. 3 +[41] Igor Santesteban, Nils Thuerey, Miguel A Otaduy, and Dan Casas. Self-supervised collision handling via generative 3d garment models for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11763-11773, 2021. 2 +[42] Igor Santesteban, Miguel Otaduy, Nils Thuerey, and Dan Casas. Ulnef: Untangled layered neural fields for mix-and-match virtual try-on. Advances in Neural Information Processing Systems, 35:12110-12125, 2022. 2 +[43] Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International Conference on Machine Learning, pages 2256-2265. PMLR, 2015. 3 +[44] Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020.3 +[45] Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. Advances in Neural Information Processing Systems, 32, 2019. 3 +[46] Haoyu Wang, Zhilu Zhang, Donglin Di, Shiliang Zhang, and Wangmeng Zuo. Mv-vton: Multi-view virtual try-on with diffusion models. arXiv preprint arXiv:2404.17364, 2024. 7 +[47] Junjie Wang, Jiemin Fang, Xiaopeng Zhang, Lingxi Xie, and Qi Tian. Gaussianeditor: Editing 3d gaussians delicately with text instructions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20902-20911, 2024. 3 +[48] Jing Wu, Jia-Wang Bian, Xinghui Li, Guangrun Wang, Ian Reid, Philip Torr, and Victor Adrian Prisacariu. Gaussctrl: multi-view consistent text-driven 3d gaussian splattering editing. arXiv preprint arXiv:2403.08733, 2024. 3, 6, 7 +[49] Jay Zhangjie Wu, Yixiao Ge, Xintao Wang, Stan Weixian Lei, Yuchao Gu, Yufei Shi, Wynne Hsu, Ying Shan, Xiaohu Qie, and Mike Zheng Shou. Tune-a-video: One-shot tuning of image diffusion models for text-to-video generation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 7623-7633, 2023. 5 +[50] Zhenyu Xie, Haoye Dong, Yufei Gao, Zehua Ma, and Xiaodan Liang. Dreamvton: Customizing 3d virtual try-on with personalized diffusion models. arXiv preprint arXiv:2407.16511, 2024. 2, 3 +[51] Zhangyang Xiong, Chenghong Li, Kenkun Liu, Hongjie Liao, Jianqiao Hu, Junyi Zhu, Shuliang Ning, Lingteng Qiu, + +Chongjie Wang, Shijie Wang, et al. Mvhumannet: A large-scale dataset of multi-view daily dressing human captures. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19801-19811, 2024. 2, 6, 7 +[52] Dejia Xu, Weili Nie, Chao Liu, Sifei Liu, Jan Kautz, Zhangyang Wang, and Arash Vahdat. Camco: Camera-controllable 3d-consistent image-to-video generation. arXiv preprint arXiv:2406.02509, 2024. 5 +[53] Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen. Oot-diffusion: Outfitting fusion based latent diffusion for controllable virtual try-on. arXiv preprint arXiv:2403.01779, 2024. 3, 4, 8 +[54] Han Yang, Xinrui Yu, and Ziwei Liu. Full-range virtual try-on with recurrent tri-level transform. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3460-3469, 2022. 2 +[55] Tao Yu, Zerong Zheng, Kaiwen Guo, Pengpeng Liu, Qionghai Dai, and Yebin Liu. Function4d: Real-time human volumetric capture from very sparse consumer rgbd sensors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5746-5756, 2021. 2, 6, 7, 8 +[56] Jinsong Zhang, Kun Li, Yu-Kun Lai, and Jingyu Yang. Pise: Person image synthesis and editing with decoupled gan. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7982-7990, 2021. 2 +[57] Lvmin Zhang and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. arXiv preprint arXiv:2302.05543, 2023. 3 +[58] Yabo Zhang, Yuxiang Wei, Dongsheng Jiang, Xiaopeng Zhang, Wangmeng Zuo, and Qi Tian. Controlvideo: Training-free controllable text-to-video generation. arXiv preprint arXiv:2305.13077, 2023. 5 +[59] Fuwei Zhao, Zhenyu Xie, Michael Kampffmeyer, Haoye Dong, Songfang Han, Tianxiang Zheng, Tao Zhang, and Xiaodan Liang. M3d-vton: A monocular-to-3d virtual try-on network. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 13239-13249, 2021. 3 +[60] Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Sahara, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two unets. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4606-4615, 2023. 2 +[61] Jingyu Zhuang, Chen Wang, Liang Lin, Lingjie Liu, and Guanbin Li. Dreameditor: Text-driven 3d scene editing with neural fields. In SIGGRAPH Asia 2023 Conference Papers, pages 1-10, 2023. 3 +[62] Jingyu Zhuang, Di Kang, Linchao Bao, Liang Lin, and Guanbin Li. Dagensm: Disentangled avatar generation with gs-enhanced mesh. arXiv preprint arXiv:2411.15205, 2024.3 +[63] Jingyu Zhuang, Di Kang, Yan-Pei Cao, Guanbin Li, Liang Lin, and Ying Shan. Tip-editor: An accurate 3d editor following both text-prompts and image-prompts. arXiv preprint arXiv:2401.14828, 2024. 3, 6, 7, 1 \ No newline at end of file diff --git a/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/images.zip b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/images.zip new file mode 100644 index 0000000000000000000000000000000000000000..d914fdc32a59ca7ce2ff8c2bc5209a0fb2693d94 --- /dev/null +++ b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/images.zip @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e50aebfd5dd0c87141ae553f24ad8a64757c81dea9d2b42631a40a528571e483 +size 650200 diff --git a/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/layout.json b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/layout.json new file mode 100644 index 0000000000000000000000000000000000000000..d6a604d2c44828cecd177056efbe07a92ba0f945 --- /dev/null +++ b/2025/VTON 360_ High-Fidelity Virtual Try-On from Any Viewing Direction/layout.json @@ -0,0 +1,8324 @@ +{ + "pdf_info": [ + { + "para_blocks": [ + { + "bbox": [ + 87, + 103, + 522, + 121 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 87, + 103, + 522, + 121 + ], + "spans": [ + { + "bbox": [ + 87, + 103, + 522, + 121 + ], + "type": "text", + "content": "VTON 360: High-Fidelity Virtual Try-On from Any Viewing Direction" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 61, + 126, + 550, + 198 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 61, + 126, + 550, + 198 + ], + "spans": [ + { + "bbox": [ + 61, + 126, + 550, + 198 + ], + "type": "text", + "content": "Zijian He1 Yuwei Ning1 Yipeng Qin2 Guangrun Wang1 Sibei Yang3 Liang Lin1,4,5 Guanbin Li1,4,5* \n1Sun Yat-sen University \n2Cardiff University \n3ShanghaiTech University \n4Guangdong Key Laboratory of Big Data Analysis and Processing \n5Peng Cheng Laboratory \nhezj39@mail2.sysu.edu.cn, yuwei.ning@hust.edu.cn, {qinyipengl991, wanggrun}@gmail.com, yangsb@shanghaiitech.edu.cn linliang@ieee.org, liguanbin@mail.sysu.edu.cn" + } + ] + } + ], + "index": 3 + }, + { + "type": "image", + "bbox": [ + 56, + 200, + 312, + 376 + ], + "blocks": [ + { + "bbox": [ + 56, + 200, + 312, + 376 + ], + "lines": [ + { + "bbox": [ + 56, + 200, + 312, + 376 + ], + "spans": [ + { + "bbox": [ + 56, + 200, + 312, + 376 + ], + "type": "image", + "image_path": "3e4ed0ca9fe9c0466bbb1e4f4019077b1eea386c4b063ea02f3897e6315ec9c9.jpg" + } + ] + } + ], + "index": 4, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 54, + 380, + 555, + 427 + ], + "lines": [ + { + "bbox": [ + 54, + 380, + 555, + 427 + ], + "spans": [ + { + "bbox": [ + 54, + 380, + 555, + 427 + ], + "type": "text", + "content": "Figure 1. Results of VTON 360. Our VTON 360 enables high-fidelity 3D Virtual Try-On (VTON) by seamlessly adapting E-commerce garments onto a clothed 3D human model, supporting full " + }, + { + "bbox": [ + 54, + 380, + 555, + 427 + ], + "type": "inline_equation", + "content": "360^{\\circ}" + }, + { + "bbox": [ + 54, + 380, + 555, + 427 + ], + "type": "text", + "content": " view rendering. The highlighted bounding boxes (dashed line) demonstrate our method's ability to preserve intricate clothing details and patterns (e.g., collar accessories, horizontal line patterns, logos, texts, numbers) across diverse garment types." + } + ] + } + ], + "index": 7, + "angle": 0, + "type": "image_caption" + } + ], + "index": 4 + }, + { + "type": "image", + "bbox": [ + 317, + 200, + 434, + 376 + ], + "blocks": [ + { + "bbox": [ + 317, + 200, + 434, + 376 + ], + "lines": [ + { + "bbox": [ + 317, + 200, + 434, + 376 + ], + "spans": [ + { + "bbox": [ + 317, + 200, + 434, + 376 + ], + "type": "image", + "image_path": "18ca224d94844f9dde5541e3e34a54b63839bcf1d69569779a98633de60a2a1f.jpg" + } + ] + } + ], + "index": 5, + "angle": 0, + "type": "image_body" + } + ], + "index": 5 + }, + { + "type": "image", + "bbox": [ + 437, + 200, + 555, + 376 + ], + "blocks": [ + { + "bbox": [ + 437, + 200, + 555, + 376 + ], + "lines": [ + { + "bbox": [ + 437, + 200, + 555, + 376 + ], + "spans": [ + { + "bbox": [ + 437, + 200, + 555, + 376 + ], + "type": "image", + "image_path": "a5889a307af41522f81b308115e028b9c3f02dce5c03771e67bcdd87075b042d.jpg" + } + ] + } + ], + "index": 6, + "angle": 0, + "type": "image_body" + } + ], + "index": 6 + }, + { + "bbox": [ + 151, + 430, + 200, + 442 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 151, + 430, + 200, + 442 + ], + "spans": [ + { + "bbox": [ + 151, + 430, + 200, + 442 + ], + "type": "text", + "content": "Abstract" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 54, + 451, + 297, + 691 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 451, + 297, + 691 + ], + "spans": [ + { + "bbox": [ + 54, + 451, + 297, + 691 + ], + "type": "text", + "content": "Virtual Try-On (VTON) is a transformative technology in e-commerce and fashion design, enabling realistic digital visualization of clothing on individuals. In this work, we propose VTON 360, a novel 3D VTON method that addresses the open challenge of achieving high-fidelity VTON that supports any-view rendering. Specifically, we leverage the equivalence between a 3D model and its rendered multiview 2D images, and reformulate 3D VTON as an extension of 2D VTON that ensures 3D consistent results across multiple views. To achieve this, we extend 2D VTON models to include multi-view garments and clothing-agnostic human body images as input, and propose several novel techniques to enhance them, including: i) a pseudo-3D pose representation using normal maps derived from the SMPL-X 3D human model, ii) a multi-view spatial attention mechanism that models the correlations between features from different viewing angles, and iii) a multi-view CLIP embedding that enhances the garment CLIP features used in 2D VTON with camera information. Extensive experiments on large-scale real datasets and clothing images from e-commerce" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 431, + 555, + 467 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 431, + 555, + 467 + ], + "spans": [ + { + "bbox": [ + 313, + 431, + 555, + 467 + ], + "type": "text", + "content": "platforms demonstrate the effectiveness of our approach. \nProject page: https://scnuhealthy.github.io/ \nVTON360." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 314, + 476, + 394, + 488 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 476, + 394, + 488 + ], + "spans": [ + { + "bbox": [ + 314, + 476, + 394, + 488 + ], + "type": "text", + "content": "1. Introduction" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 312, + 497, + 555, + 593 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 312, + 497, + 555, + 593 + ], + "spans": [ + { + "bbox": [ + 312, + 497, + 555, + 593 + ], + "type": "text", + "content": "Virtual Try-On (VTON) enables realistic digital visualization of clothing on individuals and has emerged as a transformative technology in e-commerce and fashion design. While significant research efforts have been made on 2D VTON solutions [8, 12, 19, 33, 39], these approaches are inherently limited in their representation of view-related features. To overcome this limitation and enable high-fidelity any-view rendering, 3D VTON methods were introduced." + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 312, + 594, + 556, + 714 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 312, + 594, + 556, + 714 + ], + "spans": [ + { + "bbox": [ + 312, + 594, + 556, + 714 + ], + "type": "text", + "content": "3D VTON requires accurate garment transfer onto a 3D human body while ensuring realistic garment fitting, texture preservation, and 3D consistency. The two primary aims of 3D VTON are i) achieving high-fidelity and ii) supporting any-view rendering. Leveraging the inherent capability of 3D models for any-view rendering, early 3D VTON methods [13, 15, 28] make clothing simulation on synthetic human bodies. Specifically, these methods utilized 3D scanners to capture clothing meshes, followed by the development of specialized dressing algorithms. Although effec" + } + ] + } + ], + "index": 13 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "spans": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "type": "text", + "content": "CVF" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 145, + 0, + 494, + 37 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 145, + 0, + 494, + 37 + ], + "spans": [ + { + "bbox": [ + 145, + 0, + 494, + 37 + ], + "type": "text", + "content": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 67, + 702, + 190, + 712 + ], + "type": "page_footnote", + "angle": 0, + "lines": [ + { + "bbox": [ + 67, + 702, + 190, + 712 + ], + "spans": [ + { + "bbox": [ + 67, + 702, + 190, + 712 + ], + "type": "text", + "content": "* Corresponding author is Guanbin Li." + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "26388" + } + ] + } + ], + "index": 15 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 0 + }, + { + "para_blocks": [ + { + "bbox": [ + 55, + 72, + 294, + 240 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 72, + 294, + 240 + ], + "spans": [ + { + "bbox": [ + 55, + 72, + 294, + 240 + ], + "type": "text", + "content": "tive, these methods rely on costly 3D scanning equipment and the physical presence of the human body/clothing (i.e., not fully virtual), restricting their practicality in real-world applications. As a byproduct, most early methods focused on developing geometrically correct dressing algorithms using standard templates of human body and clothing models. Addressing this limitation, researchers extended 3D VTON by introducing algorithms that reconstruct 3D clothing models from input images, enabling the use of image-based clothing inputs [3, 32, 41, 42]. However, since input clothing images (usually frontal) are inherently 2D and lack multi-view information, this approach struggles to reconstruct high-fidelity clothing models that can be rendered well from all viewing directions." + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 243, + 295, + 567 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 243, + 295, + 567 + ], + "spans": [ + { + "bbox": [ + 56, + 243, + 295, + 567 + ], + "type": "text", + "content": "To complement this missing information, DreamVTON [50] introduces a novel approach that leverages Text-to-Image (T2I) diffusion models to reconstruct both the human body and clothing from input images. Its key insight is that T2I models learned view-agnostic \"concepts\" of both bodies and garments during their training, and that the corresponding concepts for the input body and clothing images can be obtained using LoRA [21]. By utilizing Score Distillation Sampling (SDS) [37], DreamVTON can generate visual-pleasing 3D VTON results by ensuring consistency between renderings from arbitrary viewpoints and the concepts. Nonetheless, DreamVTON's high flexibility comes at the cost of low fidelity. This limitation stems from the fact that the concepts learned by T2I models are semantic in nature, thus lacking 3D geometric consistency and pixel-level accuracy with respect to the input body and clothing images. Recently, a concurrent work, namely GaussianVTON [6], partially addressed this limitation by formulating 3D VTON as a 3D scene editing task, where a given 3D human model is edited using multi-view images generated by 2D VTON methods. While it significantly enhances the fidelity of the human body, the fidelity and 3D consistency of clothing remain problematic, as there are no 2D VTON methods that can generate multi-view images with 3D consistency. Therefore, to the best of our knowledge, achieving high-fidelity 3D VTON that supports any-view rendering remains an open challenge." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 55, + 570, + 295, + 714 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 570, + 295, + 714 + ], + "spans": [ + { + "bbox": [ + 55, + 570, + 295, + 714 + ], + "type": "text", + "content": "In this work, we address the above-mentioned challenge via proposing VTON 360, a novel 3D VTON method that achieves high-fidelity VTON from arbitrary viewing directions. Similar to GaussianVTON [6], our method edits a given 3D human model by inpainting the rendered images using a latent diffusion model. However, we set ourselves apart through our novel garment fidelity preservation strategy that can generate high-fidelity on-body garments in all viewing directions. Specifically, we first extend both the garment and clothing-agnostic human body inputs to typical 2D VTON models to leverage multi-view information, including paired front and back view garment images as well" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 313, + 72, + 553, + 382 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 72, + 553, + 382 + ], + "spans": [ + { + "bbox": [ + 313, + 72, + 553, + 382 + ], + "type": "text", + "content": "as a set of multi-view clothing-agnostic human body images sampled from random azimuth angles. Then, we propose several novel enhancements to bridge the gap between typical 2D VTON methods and our multi-view 3D consistency requirements: i) We propose a pseudo-3D pose representation using normal maps derived from the SMPL-X 3D human model, which captures fine-grained surface orientation details and provides more consistent geometry across views compared to the 2D pose representations (semantic segmentation maps) used in 2D VTON models. ii) We design a Multi-view Spatial Attention mechanism that models the correlations between features from different viewing angles, featuring a novel \"correlation\" matrix modeling the relationships among different input views. iii) We propose a multi-view CLIP embedding that enhances the garment CLIP embedding used in 2D VTON methods with camera information, thereby facilitating network learning of features relevant to a particular view. Together, these innovations enable our 2D VTON model to generate high-quality, multi-view and 3D-consistent virtual try-on results. Extensive experiments on Thuman2.0 [55] and MVHumanNet [51] datasets demonstrate that our method achieves high fidelity 3D VTON which supports any-view rendering. In addition, we show the effectiveness and generalizability of our methodology by testing it using garments from e-commerce platforms. Our conclusions include:" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 314, + 383, + 553, + 563 + ], + "type": "list", + "angle": 0, + "index": 7, + "blocks": [ + { + "bbox": [ + 314, + 383, + 553, + 418 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 383, + 553, + 418 + ], + "spans": [ + { + "bbox": [ + 314, + 383, + 553, + 418 + ], + "type": "text", + "content": "- We propose a novel 3D Virtual Try-On (VTON) method, namely VTON 360, which achieves high-fidelity VTON from arbitrary viewing directions." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 314, + 419, + 553, + 525 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 419, + 553, + 525 + ], + "spans": [ + { + "bbox": [ + 314, + 419, + 553, + 525 + ], + "type": "text", + "content": "- Leveraging the equivalence between a 3D model and its rendered multi-view 2D images, we reformulate 3D VTON as an extension of 2D VTON that ensures consistent results across multiple views. Specifically, we introduce several novel techniques, including: (i) pseudo-3D pose representation; (ii) multi-view spatial attention; and (iii) multi-view CLIP embedding. These innovations enhance traditional 2D VTON models to generate multiview and 3D-consistent results." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 314, + 526, + 553, + 563 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 526, + 553, + 563 + ], + "spans": [ + { + "bbox": [ + 314, + 526, + 553, + 563 + ], + "type": "text", + "content": "- Extensive experimental results on two large real datasets as well as real clothing images from e-commerce platforms demonstrate the effectiveness of our approach." + } + ] + } + ], + "index": 6 + } + ], + "sub_type": "text" + }, + { + "bbox": [ + 314, + 573, + 400, + 586 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 573, + 400, + 586 + ], + "spans": [ + { + "bbox": [ + 314, + 573, + 400, + 586 + ], + "type": "text", + "content": "2. Related Work" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 313, + 594, + 553, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 594, + 553, + 713 + ], + "spans": [ + { + "bbox": [ + 313, + 594, + 553, + 713 + ], + "type": "text", + "content": "2D Virtual Try-On. 2D Virtual Try-On (VTON) aims to transfer a desired garment to the corresponding region of a target human image while preserving the human pose and identity. Early methods [2, 8, 10, 11, 16, 18, 29, 31, 38, 54, 56] use Generative Adversarial Networks (GANs) to deform the garments to match the target body shape, which a critical step for achieving realistic VTON. However, accurately adapting to diverse real-world conditions remains a significant challenge. Addressing this issue, recent methods [12, 19, 33, 60] reframe 2D VTON as a conditioned in-" + } + ] + } + ], + "index": 9 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "26389" + } + ] + } + ], + "index": 10 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 1 + }, + { + "para_blocks": [ + { + "bbox": [ + 55, + 72, + 294, + 133 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 72, + 294, + 133 + ], + "spans": [ + { + "bbox": [ + 55, + 72, + 294, + 133 + ], + "type": "text", + "content": "painting task, leveraging the strong priors provided by diffusion models [20, 43, 45] to achieve promising results. This strategy is further improved by [9, 26, 53], which introduce a ReferenceNet to extract hierarchical garment features and apply attention mechanisms to condition the Main UNet." + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 141, + 296, + 561 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 141, + 296, + 561 + ], + "spans": [ + { + "bbox": [ + 56, + 141, + 296, + 561 + ], + "type": "text", + "content": "3D Virutal Try-On. For 3D Virtual Try-On (VTON), traditional methods [4, 13, 15, 28, 36] rely on 3D scanning or cloth simulation to generate highly precise body and garment geometry. These methods were then extended by learning-based methods [3, 32] that employ differentiable rendering to dress the SMPL [30] model with a desired garment mesh. Despite their effectiveness, such methods rely on costly 3D scanning and the physical presence of human body/clothing, limiting their application in the real world. Addressing this limitation, M3D-VTON [59] proposes a depth-based 3D VTON framework to reconstruct 3D clothed human models from 2D human and garment images, but the results often suffer from explicit warping artifacts. To improve 3D VTON results, recent methods [23, 24, 50, 62] resort to text-to-image (T2I) diffusion models and employ the Score Distillation Sampling (SDS) loss [40] to ensure consistency among different viewing directions. Specifically, TeCH [24] adapts the generative priors of T2I diffusion model to the specific person and clothes by training descriptive text prompts with DreamBooth [40]. DreamWaltz [23] leverages Pose ControlNet [57] to attain clothed human body models. DreamVTON [50] introduces a multi-concept LoRA [21] to personalize the T2I diffusion model, and uses a template-based optimization mechanism that combines with SDS loss to better preserve patterns on the garment. Although effective, these methods often produce results lacking in fidelity, as the concepts learned by T2I models are semantic rather than at the pixel level. Concurrent to our work, GaussianVTON [6] proposes an alternative approach by combining Gaussian Splatting [25] with pre-trained 2D VTON models and formulate it as an editing task. However, since there are no 2D VTON methods that can generate multi-view images with 3D consistency, the fidelity and 3D consistency of the clothing generated remain problematic." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 55, + 570, + 295, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 570, + 295, + 713 + ], + "spans": [ + { + "bbox": [ + 55, + 570, + 295, + 713 + ], + "type": "text", + "content": "Radiance Field-based 3D Human or Scene Editing. Recently, radiance field-based editing has attracted significant interest due to its efficient differentiable rendering capabilities, sparking substantial advancements in text-driven 3D editing. For example, InstructN2N [17] employ an image-based diffusion model InstructP2P [5] to modify the rendered image by the user's text description with a variant of the score distillation sampling (SDS) [37] loss. GaussianEditor [7] applies Gaussian Splatting [25] as 3D representation instead of NeRF, adopting Gaussian semantic tracking to track target Gaussian values, significantly improving editing speed and controllability. To enable accu" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 313, + 72, + 555, + 228 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 72, + 555, + 228 + ], + "spans": [ + { + "bbox": [ + 313, + 72, + 555, + 228 + ], + "type": "text", + "content": "rate location and appearance control, subsequent works [47, 61] specify the editing region using the attention score or with a segmentation model. TIP-Editor [63] proposes a content personalization step dedicated to the reference image based on LoRA, achieving the editing following a hybrid text-image prompt. GaussCtrl [48] leverage depth conditions and attention-based latent code alignment to achieve 3D-aware multi-view consistent editing instead of iteratively editing single views using SDS loss. However, these works primarily focus on global appearance modifications within a text-driven pipeline, while our approach emphasizes preserving fine textural details from different viewing directions throughout the editing process." + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 313, + 238, + 391, + 251 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 238, + 391, + 251 + ], + "spans": [ + { + "bbox": [ + 313, + 238, + 391, + 251 + ], + "type": "text", + "content": "3. Preliminary" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 313, + 258, + 555, + 354 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 258, + 555, + 354 + ], + "spans": [ + { + "bbox": [ + 313, + 258, + 555, + 354 + ], + "type": "text", + "content": "Latent Diffusion Model. Latent Diffusion Model [39] is a variant of diffusion models that performs denoising within the latent space of a Variational Autoencoder (VAE) [27]. During training, given a fixed encoder " + }, + { + "bbox": [ + 313, + 258, + 555, + 354 + ], + "type": "inline_equation", + "content": "\\mathcal{E}" + }, + { + "bbox": [ + 313, + 258, + 555, + 354 + ], + "type": "text", + "content": ", an input image " + }, + { + "bbox": [ + 313, + 258, + 555, + 354 + ], + "type": "inline_equation", + "content": "x" + }, + { + "bbox": [ + 313, + 258, + 555, + 354 + ], + "type": "text", + "content": " is transformed into its latent representation " + }, + { + "bbox": [ + 313, + 258, + 555, + 354 + ], + "type": "inline_equation", + "content": "z_0 = \\mathcal{E}(x)" + }, + { + "bbox": [ + 313, + 258, + 555, + 354 + ], + "type": "text", + "content": ". A conditional diffusion model " + }, + { + "bbox": [ + 313, + 258, + 555, + 354 + ], + "type": "inline_equation", + "content": "\\hat{\\epsilon}_{\\theta}" + }, + { + "bbox": [ + 313, + 258, + 555, + 354 + ], + "type": "text", + "content": ", typically implemented with a UNet architecture, is then trained using a weighted denoising score matching objective:" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 357, + 361, + 555, + 376 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 357, + 361, + 555, + 376 + ], + "spans": [ + { + "bbox": [ + 357, + 361, + 555, + 376 + ], + "type": "interline_equation", + "content": "\\mathcal {L} _ {L D M} = \\mathbb {E} _ {\\mathbf {z}, \\mathbf {c}, \\boldsymbol {\\epsilon}, t} [ \\boldsymbol {\\epsilon} - \\| \\hat {\\epsilon} _ {\\theta} (\\mathbf {z} _ {\\mathbf {t}}; \\mathbf {c}, t) \\| _ {2} ^ {2} ] \\tag {1}", + "image_path": "0b097be8d063981c25cde6c0c31c2065a1227e4824aaae1f86881816781495f5.jpg" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "spans": [ + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "inline_equation", + "content": "\\mathbf{z}_t\\coloneqq \\alpha_t\\mathbf{x} + \\sigma_t\\boldsymbol {\\epsilon}" + }, + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "text", + "content": " denotes the forward diffusion process at timestep " + }, + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "inline_equation", + "content": "t" + }, + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "text", + "content": " . " + }, + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "inline_equation", + "content": "\\alpha_{t},\\sigma_{t}" + }, + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "text", + "content": " are time-dependent functions defined by the diffusion model formulation; c denotes the conditional input and " + }, + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "inline_equation", + "content": "\\epsilon \\sim \\mathcal{N}(\\mathbf{0},\\mathbf{1})" + }, + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "text", + "content": " is Gaussian noise. During inference, data samples are generated by initiating from Gaussian noise " + }, + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "inline_equation", + "content": "\\mathbf{z}_T\\sim \\mathcal{N}(\\mathbf{0},\\mathbf{1})" + }, + { + "bbox": [ + 313, + 384, + 556, + 468 + ], + "type": "text", + "content": " and iteratively refining it using a DDIM [44] sampler." + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 313, + 478, + 370, + 490 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 478, + 370, + 490 + ], + "spans": [ + { + "bbox": [ + 313, + 478, + 370, + 490 + ], + "type": "text", + "content": "4. Method" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 313, + 498, + 556, + 653 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 498, + 556, + 653 + ], + "spans": [ + { + "bbox": [ + 313, + 498, + 556, + 653 + ], + "type": "text", + "content": "Our method leverages the equivalence between a 3D model and its rendered multi-view 2D images to achieve high-fidelity, any-view 3D VTON. Specifically, as Fig. 2 shows, given an input 3D human model and a garment image, our method 1) renders the 3D model into multi-view 2D images and 2) formulates 3D VTON as a consistent, unified 2D VTON process across these rendered views; 3) By reconstructing the edited images into a 3D model using existing 3D reconstruction methods, we ensure visual coherence and precise garment alignment from any viewing angle. Among them, the second step is crucial as existing 2D VTON methods [9, 26, 53] lack 3D knowledge, preventing them from generating multi-view images with 3D consistency." + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 654, + 555, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 654, + 555, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 654, + 555, + 715 + ], + "type": "text", + "content": "To address this challenge, we propose several novel techniques (Sec. 4.2) that equip a typical 2D VTON network (Sec. 4.1), which is built on a latent diffusion model [39], with the capability to generate 3D-consistent results. We use Gaussian Splatting [25] as our 3D representation." + } + ] + } + ], + "index": 10 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "text", + "content": "26390" + } + ] + } + ], + "index": 11 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 2 + }, + { + "para_blocks": [ + { + "type": "image", + "bbox": [ + 55, + 70, + 553, + 342 + ], + "blocks": [ + { + "bbox": [ + 55, + 70, + 553, + 342 + ], + "lines": [ + { + "bbox": [ + 55, + 70, + 553, + 342 + ], + "spans": [ + { + "bbox": [ + 55, + 70, + 553, + 342 + ], + "type": "image", + "image_path": "50456d17e5ccbb76f3e583fe2b0ebac3fa363e2d03f09e2a13d08f82ca40471e.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 55, + 350, + 555, + 394 + ], + "lines": [ + { + "bbox": [ + 55, + 350, + 555, + 394 + ], + "spans": [ + { + "bbox": [ + 55, + 350, + 555, + 394 + ], + "type": "text", + "content": "Figure 2. Overview of VTON 360. Given an input 3D human model " + }, + { + "bbox": [ + 55, + 350, + 555, + 394 + ], + "type": "inline_equation", + "content": "\\mathbf{G}_{\\mathrm{src}}" + }, + { + "bbox": [ + 55, + 350, + 555, + 394 + ], + "type": "text", + "content": " and a pair of garment images " + }, + { + "bbox": [ + 55, + 350, + 555, + 394 + ], + "type": "inline_equation", + "content": "(g_{f}, g_{b})" + }, + { + "bbox": [ + 55, + 350, + 555, + 394 + ], + "type": "text", + "content": ", our method 1) renders " + }, + { + "bbox": [ + 55, + 350, + 555, + 394 + ], + "type": "inline_equation", + "content": "\\mathbf{G}_{\\mathrm{src}}" + }, + { + "bbox": [ + 55, + 350, + 555, + 394 + ], + "type": "text", + "content": " into multi-view 2D images (left) and 2) edits the rendered multi-view 2D images (middle); 3) reconstructs the edited images into a 3D model " + }, + { + "bbox": [ + 55, + 350, + 555, + 394 + ], + "type": "inline_equation", + "content": "\\mathbf{G}_{\\mathrm{VTON}}" + }, + { + "bbox": [ + 55, + 350, + 555, + 394 + ], + "type": "text", + "content": " (right). In the crucial step 2), we propose three novel techniques to equip a typical 2D VTON network with the capability to generate 3D-consistent results: 1) Pseudo-3D Pose Input, 2) Multi-view Spatial Attention, and 3) Multi-view CLIP Embedding." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "image_caption" + } + ], + "index": 0 + }, + { + "type": "image", + "bbox": [ + 56, + 399, + 294, + 538 + ], + "blocks": [ + { + "bbox": [ + 56, + 399, + 294, + 538 + ], + "lines": [ + { + "bbox": [ + 56, + 399, + 294, + 538 + ], + "spans": [ + { + "bbox": [ + 56, + 399, + 294, + 538 + ], + "type": "image", + "image_path": "c3dd68bc18957e8ae4d2aa15c28ee2931189de321d4ce4c838cc8d0a92c0acbb.jpg" + } + ] + } + ], + "index": 2, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 55, + 544, + 295, + 610 + ], + "lines": [ + { + "bbox": [ + 55, + 544, + 295, + 610 + ], + "spans": [ + { + "bbox": [ + 55, + 544, + 295, + 610 + ], + "type": "text", + "content": "Figure 3. DensePose (2D) vs. SMPL-X normal map (pseudo-3D) representations. DensePose applies uniform labels per body part, lacking 3D consistency across views and causing artifacts and temporal inconsistencies (highlighted with red boxes). In contrast, SMPL-X normal maps capture fine surface details, ensuring geometric coherence and stable, realistic shading across views." + } + ] + } + ], + "index": 3, + "angle": 0, + "type": "image_caption" + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 621, + 228, + 634 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 621, + 228, + 634 + ], + "spans": [ + { + "bbox": [ + 55, + 621, + 228, + 634 + ], + "type": "text", + "content": "4.1. Recap of 2D VTON Framework" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 641, + 296, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 641, + 296, + 713 + ], + "spans": [ + { + "bbox": [ + 55, + 641, + 296, + 713 + ], + "type": "text", + "content": "Following previous works [12, 26, 53], we formulate 2D VTON as an exemplar-based inpainting problem, aiming to fill an input clothing-agnostic image " + }, + { + "bbox": [ + 55, + 641, + 296, + 713 + ], + "type": "inline_equation", + "content": "\\mathbf{A}" + }, + { + "bbox": [ + 55, + 641, + 296, + 713 + ], + "type": "text", + "content": " with a given garment image " + }, + { + "bbox": [ + 55, + 641, + 296, + 713 + ], + "type": "inline_equation", + "content": "g" + }, + { + "bbox": [ + 55, + 641, + 296, + 713 + ], + "type": "text", + "content": ", where " + }, + { + "bbox": [ + 55, + 641, + 296, + 713 + ], + "type": "inline_equation", + "content": "\\mathbf{A}" + }, + { + "bbox": [ + 55, + 641, + 296, + 713 + ], + "type": "text", + "content": " is obtained following the method used in [53]. As illustrated in Fig. 2 (middle), the network architecture is based on the latent diffusion model [39] with" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 314, + 401, + 500, + 412 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 401, + 500, + 412 + ], + "spans": [ + { + "bbox": [ + 314, + 401, + 500, + 412 + ], + "type": "text", + "content": "an encoder " + }, + { + "bbox": [ + 314, + 401, + 500, + 412 + ], + "type": "inline_equation", + "content": "\\mathcal{E}" + }, + { + "bbox": [ + 314, + 401, + 500, + 412 + ], + "type": "text", + "content": " and comprises two components:" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 313, + 412, + 553, + 507 + ], + "type": "list", + "angle": 0, + "index": 9, + "blocks": [ + { + "bbox": [ + 314, + 412, + 544, + 424 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 412, + 544, + 424 + ], + "spans": [ + { + "bbox": [ + 314, + 412, + 544, + 424 + ], + "type": "text", + "content": "- A GarmentNet [9, 53] that extracts features from " + }, + { + "bbox": [ + 314, + 412, + 544, + 424 + ], + "type": "inline_equation", + "content": "\\mathcal{E}(g)" + }, + { + "bbox": [ + 314, + 412, + 544, + 424 + ], + "type": "text", + "content": "." + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 313, + 424, + 553, + 507 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 424, + 553, + 507 + ], + "spans": [ + { + "bbox": [ + 313, + 424, + 553, + 507 + ], + "type": "text", + "content": "- A Main UNet that inpaints " + }, + { + "bbox": [ + 313, + 424, + 553, + 507 + ], + "type": "inline_equation", + "content": "\\mathbf{A}" + }, + { + "bbox": [ + 313, + 424, + 553, + 507 + ], + "type": "text", + "content": " according to i) detailed garment features extracted by the GarmentNet; ii) the 2D pose of " + }, + { + "bbox": [ + 313, + 424, + 553, + 507 + ], + "type": "inline_equation", + "content": "\\mathbf{A}" + }, + { + "bbox": [ + 313, + 424, + 553, + 507 + ], + "type": "text", + "content": " represented by semantic labels using Dense-Pose [14]; iii) CLIP embeddings of input garment " + }, + { + "bbox": [ + 313, + 424, + 553, + 507 + ], + "type": "inline_equation", + "content": "g" + }, + { + "bbox": [ + 313, + 424, + 553, + 507 + ], + "type": "text", + "content": ". Among them, i) and ii) together with noise are input to the self-attention layers, while iii) is input to the cross-attention layers of the Main UNet." + } + ] + } + ], + "index": 8 + } + ], + "sub_type": "text" + }, + { + "bbox": [ + 313, + 508, + 553, + 531 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 508, + 553, + 531 + ], + "spans": [ + { + "bbox": [ + 313, + 508, + 553, + 531 + ], + "type": "text", + "content": "Both the GarmentNet and the Main UNet share the same network architecture." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 313, + 540, + 537, + 553 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 540, + 537, + 553 + ], + "spans": [ + { + "bbox": [ + 313, + 540, + 537, + 553 + ], + "type": "text", + "content": "4.2. Multi-view 2D VTON with 3D Consistency" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 313, + 558, + 553, + 594 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 558, + 553, + 594 + ], + "spans": [ + { + "bbox": [ + 313, + 558, + 553, + 594 + ], + "type": "text", + "content": "To enable the aforementioned 2D VTON model to generate multi-view and 3D-consistent results, we propose the following novel enhancements to its design:" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 314, + 594, + 543, + 605 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 594, + 543, + 605 + ], + "spans": [ + { + "bbox": [ + 314, + 594, + 543, + 605 + ], + "type": "text", + "content": "Multi-view Inputs. We extend both inputs to the model:" + } + ] + } + ], + "index": 13 + }, + { + "bbox": [ + 314, + 606, + 554, + 714 + ], + "type": "list", + "angle": 0, + "index": 16, + "blocks": [ + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "spans": [ + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "text", + "content": "- Multi-view Garment Inputs: We extend the input garment representation from a single image " + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "inline_equation", + "content": "g" + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "text", + "content": " to paired front and back view images " + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "inline_equation", + "content": "g_{f}" + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "text", + "content": ", " + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "inline_equation", + "content": "g_{b}" + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "text", + "content": ", providing comprehensive garment information across all viewing angles. Accordingly, we use the encoder " + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "inline_equation", + "content": "\\mathcal{E}" + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "text", + "content": " to encode " + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "inline_equation", + "content": "g_{f}, g_{b}" + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "text", + "content": " into their latent representations " + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "inline_equation", + "content": "\\mathcal{E}(g_{f}), \\mathcal{E}(g_{b})" + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "text", + "content": " and feed them into GarmentNet to obtain their multi-layer features " + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "inline_equation", + "content": "F_{f}^{l}" + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "inline_equation", + "content": "F_{b}^{l}" + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "text", + "content": " at layer " + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "inline_equation", + "content": "l" + }, + { + "bbox": [ + 315, + 606, + 554, + 701 + ], + "type": "text", + "content": ", respectively." + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 314, + 701, + 553, + 714 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 701, + 553, + 714 + ], + "spans": [ + { + "bbox": [ + 314, + 701, + 553, + 714 + ], + "type": "text", + "content": "- Multi-view Clothing-agnostic Image Inputs: Based on" + } + ] + } + ], + "index": 15 + } + ], + "sub_type": "text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "26391" + } + ] + } + ], + "index": 17 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 3 + }, + { + "para_blocks": [ + { + "bbox": [ + 63, + 72, + 295, + 167 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 63, + 72, + 295, + 167 + ], + "spans": [ + { + "bbox": [ + 63, + 72, + 295, + 167 + ], + "type": "text", + "content": "the equivalence between a 3D human model and its rendered multi-view 2D images, we extend the input human body representation from a single, clothing-agnostic image, " + }, + { + "bbox": [ + 63, + 72, + 295, + 167 + ], + "type": "inline_equation", + "content": "\\mathbf{A}" + }, + { + "bbox": [ + 63, + 72, + 295, + 167 + ], + "type": "text", + "content": ", to a set of " + }, + { + "bbox": [ + 63, + 72, + 295, + 167 + ], + "type": "inline_equation", + "content": "m" + }, + { + "bbox": [ + 63, + 72, + 295, + 167 + ], + "type": "text", + "content": " multi-view images, denoted as " + }, + { + "bbox": [ + 63, + 72, + 295, + 167 + ], + "type": "inline_equation", + "content": "\\mathbf{A}_1, \\mathbf{A}_2, \\dots, \\mathbf{A}_m" + }, + { + "bbox": [ + 63, + 72, + 295, + 167 + ], + "type": "text", + "content": ". These images are sampled from random azimuth angles, allowing the 2D VTON model to access comprehensive, multi-view information from the input 3D human model." + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 168, + 295, + 372 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 168, + 295, + 372 + ], + "spans": [ + { + "bbox": [ + 55, + 168, + 295, + 372 + ], + "type": "text", + "content": "Pseudo-3D Pose Input. As shown in Fig. 3, the 2D Dense-Pose representations [14] commonly used in state-of-the-art 2D VTON methods [9, 26] assign a uniform semantic label to all pixels within each body part (e.g., thigh), inherently lack 3D geometric consistency across multiple views, and often introduce artifacts and temporal inconsistencies. To address these limitations, we propose a novel pseudo-3D pose representation: the normal maps " + }, + { + "bbox": [ + 55, + 168, + 295, + 372 + ], + "type": "inline_equation", + "content": "\\mathbf{N}" + }, + { + "bbox": [ + 55, + 168, + 295, + 372 + ], + "type": "text", + "content": " derived from the SMPL-X [35] model of the input body. These normal maps capture fine-grained surface orientation details, providing a more consistent representation across views by preserving geometric structure in the 3D space. Furthermore, they enable smoother, temporally stable transitions and realistic shading effects across multi-view scenarios. In practice, we employ a lightweight PoseEncoder " + }, + { + "bbox": [ + 55, + 168, + 295, + 372 + ], + "type": "inline_equation", + "content": "\\mathcal{E}'" + }, + { + "bbox": [ + 55, + 168, + 295, + 372 + ], + "type": "text", + "content": " [22] and feed " + }, + { + "bbox": [ + 55, + 168, + 295, + 372 + ], + "type": "inline_equation", + "content": "\\mathcal{E}'(\\mathbf{N})" + }, + { + "bbox": [ + 55, + 168, + 295, + 372 + ], + "type": "text", + "content": " into the Main UNet. We obtain the SMPL-X model from the multi-view images of the input body using EasyMoCap [1]." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "spans": [ + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "text", + "content": "Accordingly, we concatenate three components as the enhanced input to the Main UNet: i) a noise latent " + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "inline_equation", + "content": "z_{t}" + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "text", + "content": "; ii) the encoded pseudo-3D poses " + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "inline_equation", + "content": "\\mathcal{E}'(\\mathbf{N_1}),\\mathcal{E}'(\\mathbf{N_2}),\\dots,\\mathcal{E}'(\\mathbf{N_m})" + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "text", + "content": "; and iii) the encoded multi-view clothing-agnostic images " + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "inline_equation", + "content": "\\mathcal{E}(\\mathbf{A_1}),\\mathcal{E}(\\mathbf{A_2}),\\dots,\\mathcal{E}(\\mathbf{A_m})" + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "text", + "content": ". Let " + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "inline_equation", + "content": "F_{1}^{l},F_{2}^{l},\\dots,F_{m}^{l}" + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "text", + "content": " be the feature representations at layer " + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "inline_equation", + "content": "l" + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "text", + "content": " of the Main UNet, and recall the garment features " + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "inline_equation", + "content": "F_{f}^{l}" + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "inline_equation", + "content": "F_{b}^{l}" + }, + { + "bbox": [ + 55, + 373, + 295, + 467 + ], + "type": "text", + "content": " defined above, we enhance the self-attention layers of the Main UNet as:" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 468, + 295, + 647 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 468, + 295, + 647 + ], + "spans": [ + { + "bbox": [ + 55, + 468, + 295, + 647 + ], + "type": "text", + "content": "Multi-view Spatial Attention. To cope with the aforementioned multi-view input features and ensure their consistency, we draw insights from the temporal attention layer commonly used in video generation and editing [49, 58] and extend it to our multi-view spatial attention layer, denoted as MVAttention. The key distinction of our MVAttention is that its input multi-view features " + }, + { + "bbox": [ + 55, + 468, + 295, + 647 + ], + "type": "inline_equation", + "content": "F_{1}^{l}, F_{2}^{l}, \\ldots, F_{m}^{l}" + }, + { + "bbox": [ + 55, + 468, + 295, + 647 + ], + "type": "text", + "content": " are from images captured from non-uniform spatial intervals, with the viewing angles varying randomly. Consequently, features from similar views exhibit a higher correlation, while those from distinct views are largely independent. To model this relationship, we construct a \"correlation\" matrix " + }, + { + "bbox": [ + 55, + 468, + 295, + 647 + ], + "type": "inline_equation", + "content": "C" + }, + { + "bbox": [ + 55, + 468, + 295, + 647 + ], + "type": "text", + "content": " based on the angular disparity obtained from camera rotation matrices of the input multi-view images, and define our MVAttention as follows:" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 69, + 654, + 294, + 717 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 69, + 654, + 294, + 717 + ], + "spans": [ + { + "bbox": [ + 69, + 654, + 294, + 717 + ], + "type": "interline_equation", + "content": "\\begin{array}{l} \\mathbf {F} ^ {\\mathbf {l}} = \\left[ F _ {1} ^ {l} \\oplus F _ {2} ^ {l} \\dots \\oplus F _ {m} ^ {l} \\right], \\hat {\\mathbf {F}} ^ {\\mathbf {l}} = \\left[ \\mathbf {F} ^ {\\mathbf {l}} \\oplus F _ {f} ^ {l} \\oplus F _ {b} ^ {l} \\right] \\\\ Q = W ^ {Q} \\mathbf {F} ^ {1}, K = W ^ {K} \\hat {\\mathbf {F}} ^ {1}, V = W ^ {V} \\hat {\\mathbf {F}} ^ {1} \\tag {2} \\\\ A _ {i} = \\operatorname {s o f t m a x} \\left(\\frac {Q _ {i} \\times \\left(C _ {i} \\cdot K ^ {T}\\right)}{\\sqrt {d}}\\right), \\mathbf {H} _ {\\mathbf {i}} ^ {\\mathbf {l}} = A _ {i} \\times V _ {i} \\\\ \\end{array}", + "image_path": "50aba07810aab3ae41365a6cc5de1abfb0f696fc8cce167f92acff6605cef86f.jpg" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "spans": [ + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "i\\in \\{1,2,\\dots,m\\}" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " denotes " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "i" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " -th view; the Query " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "Q" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " comes directly from " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "\\mathbf{F}^1" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " and the concatenation of " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "[\\mathbf{F}^1,F_f^l,F_b^l ]" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " serves as the key " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "K" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " and the value " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "V" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " . " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "\\oplus" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " indicates matrix concatenation along the token axis; " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "d" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " denotes the dimension; " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "W^{Q},W^{K},W^{V}" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " represent the linear transformation matrices; we omitted the " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "l" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " of the attention matrices and parameters for simplicity; " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "C\\in \\mathbb{R}^{m\\times m}" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "C_i" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " represents " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "i" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " -th row in " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "C" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " , and its \"correlation\" value between " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "i" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " -th and " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "j" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " -th features is " + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "inline_equation", + "content": "C_{ij}" + }, + { + "bbox": [ + 313, + 72, + 553, + 180 + ], + "type": "text", + "content": " .." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 357, + 190, + 553, + 205 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 357, + 190, + 553, + 205 + ], + "spans": [ + { + "bbox": [ + 357, + 190, + 553, + 205 + ], + "type": "interline_equation", + "content": "C _ {i j} = \\left(\\left(\\operatorname {t r a c e} \\left(R _ {i} ^ {T} R _ {j}\\right) - 1\\right) / 2 + 1\\right) / 2 \\tag {3}", + "image_path": "e9f2c598c075ed516a39b928d28c56920869a5a2f8b1c7ee298783119273cdea.jpg" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 313, + 213, + 553, + 249 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 213, + 553, + 249 + ], + "spans": [ + { + "bbox": [ + 313, + 213, + 553, + 249 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 313, + 213, + 553, + 249 + ], + "type": "inline_equation", + "content": "R_{i}" + }, + { + "bbox": [ + 313, + 213, + 553, + 249 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 313, + 213, + 553, + 249 + ], + "type": "inline_equation", + "content": "R_{j}" + }, + { + "bbox": [ + 313, + 213, + 553, + 249 + ], + "type": "text", + "content": " are the extrinsic rotation matrices of the corresponding camera views, " + }, + { + "bbox": [ + 313, + 213, + 553, + 249 + ], + "type": "inline_equation", + "content": "(\\mathrm{trace}(R_i^T R_j) - 1) / 2" + }, + { + "bbox": [ + 313, + 213, + 553, + 249 + ], + "type": "text", + "content": " is the cosine value of the angle between these camera views." + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 313, + 250, + 554, + 418 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 250, + 554, + 418 + ], + "spans": [ + { + "bbox": [ + 313, + 250, + 554, + 418 + ], + "type": "text", + "content": "Multi-view CLIP Embedding. Camera viewpoints can serve as an effective condition signal to enhance 3D consistency in video content generation [52]. Building on this insight, we incorporate camera condition within our try-on network by encoding camera parameters as an additional token, enabling the generation of more consistent multi-view images. Specifically, we define a world coordinate system in which the camera faces the subject directly. For each input image (view) " + }, + { + "bbox": [ + 313, + 250, + 554, + 418 + ], + "type": "inline_equation", + "content": "\\mathbf{A}_{\\mathrm{i}}" + }, + { + "bbox": [ + 313, + 250, + 554, + 418 + ], + "type": "text", + "content": ", " + }, + { + "bbox": [ + 313, + 250, + 554, + 418 + ], + "type": "inline_equation", + "content": "1 \\leq i \\leq m" + }, + { + "bbox": [ + 313, + 250, + 554, + 418 + ], + "type": "text", + "content": ", we extract the rotation matrix from the camera's corresponding extrinsic matrix. This rotation matrix is then reshaped into a 9-dimensional tensor " + }, + { + "bbox": [ + 313, + 250, + 554, + 418 + ], + "type": "inline_equation", + "content": "\\mathbf{r}_{\\mathrm{i}}" + }, + { + "bbox": [ + 313, + 250, + 554, + 418 + ], + "type": "text", + "content": ", which undergoes positional encoding to effectively integrate the camera parameters into the feature representation " + }, + { + "bbox": [ + 313, + 250, + 554, + 418 + ], + "type": "inline_equation", + "content": "F_{i}^{c}" + }, + { + "bbox": [ + 313, + 250, + 554, + 418 + ], + "type": "text", + "content": "." + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 364, + 427, + 553, + 458 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 364, + 427, + 553, + 458 + ], + "spans": [ + { + "bbox": [ + 364, + 427, + 553, + 458 + ], + "type": "interline_equation", + "content": "\\begin{array}{l} F _ {i} ^ {c} = \\left(\\sin \\left(2 ^ {0} \\pi \\mathbf {r} _ {\\mathbf {i}}\\right), \\cos \\left(2 ^ {0} \\pi \\mathbf {r} _ {\\mathbf {i}}\\right), \\dots , \\right. \\tag {4} \\\\ \\sin (2 ^ {L - 1} \\pi \\mathbf {r _ {i}}), \\cos (2 ^ {L - 1} \\pi \\mathbf {r _ {i}})) \\\\ \\end{array}", + "image_path": "f64ec9f145daf570f7f3cc44754c93ada6aa47f20037f1d93d7dfbe677f07fe2.jpg" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "spans": [ + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "inline_equation", + "content": "L" + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "text", + "content": " is the length of positional embedding. We then project " + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "inline_equation", + "content": "F_{i}^{c}" + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "text", + "content": " to match the dimensionality of the garment CLIP image embedding " + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "inline_equation", + "content": "F^{g}" + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "text", + "content": " via an MLP and concatenate them along the token axis to form " + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "inline_equation", + "content": "Y_{i}" + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "text", + "content": ". This combined representation, " + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "inline_equation", + "content": "Y_{i}" + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "text", + "content": ", is subsequently used in the key " + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "inline_equation", + "content": "K_{\\mathrm{x}}" + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "text", + "content": " and value " + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "inline_equation", + "content": "V_{\\mathrm{x}}" + }, + { + "bbox": [ + 313, + 468, + 554, + 538 + ], + "type": "text", + "content": " of the cross-attention layers of the Main UNet:" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 389, + 549, + 476, + 562 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 389, + 549, + 476, + 562 + ], + "spans": [ + { + "bbox": [ + 389, + 549, + 476, + 562 + ], + "type": "interline_equation", + "content": "Y _ {i} = F ^ {g} \\oplus \\operatorname {M L P} \\left(F _ {i} ^ {c}\\right)", + "image_path": "ac33b9e09b90e4734e6a9e5489419d3ff7fa37e5a85fadebc9744ba70e68ce60.jpg" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 343, + 564, + 553, + 583 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 343, + 564, + 553, + 583 + ], + "spans": [ + { + "bbox": [ + 343, + 564, + 553, + 583 + ], + "type": "interline_equation", + "content": "Q _ {\\mathrm {x}} = W _ {\\mathrm {x}} ^ {Q} \\mathbf {H} _ {\\mathrm {i}} ^ {1}, K _ {\\mathrm {x}} = W _ {\\mathrm {x}} ^ {K} Y _ {i}, V _ {\\mathrm {x}} = W _ {\\mathrm {x}} ^ {V} Y _ {i} \\tag {5}", + "image_path": "14b922f3fee998b3c504d955a9df1e61da6a6ca47f3f7068495902f8f15ff8a7.jpg" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 359, + 582, + 486, + 609 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 359, + 582, + 486, + 609 + ], + "spans": [ + { + "bbox": [ + 359, + 582, + 486, + 609 + ], + "type": "interline_equation", + "content": "F _ {i} ^ {(l + 1)} = \\mathrm {s o f t m a x} (\\frac {Q _ {\\mathrm {x}} K _ {\\mathrm {x}} ^ {T}}{\\sqrt {d _ {\\mathrm {x}}}}) V _ {\\mathrm {x}},", + "image_path": "09d0f4d0b62b2a766fa1c039db6075c67fd6c21d36a4a32bdcfcf1e164459876.jpg" + } + ] + } + ], + "index": 13 + }, + { + "bbox": [ + 313, + 617, + 553, + 654 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 617, + 553, + 654 + ], + "spans": [ + { + "bbox": [ + 313, + 617, + 553, + 654 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 313, + 617, + 553, + 654 + ], + "type": "inline_equation", + "content": "\\mathbf{H}^1" + }, + { + "bbox": [ + 313, + 617, + 553, + 654 + ], + "type": "text", + "content": " is the output of the MVAttention of the " + }, + { + "bbox": [ + 313, + 617, + 553, + 654 + ], + "type": "inline_equation", + "content": "l" + }, + { + "bbox": [ + 313, + 617, + 553, + 654 + ], + "type": "text", + "content": "-th layer; we omitted the " + }, + { + "bbox": [ + 313, + 617, + 553, + 654 + ], + "type": "inline_equation", + "content": "l" + }, + { + "bbox": [ + 313, + 617, + 553, + 654 + ], + "type": "text", + "content": " of the cross attention matrices and parameters for simplicity." + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 313, + 654, + 553, + 689 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 654, + 553, + 689 + ], + "spans": [ + { + "bbox": [ + 313, + 654, + 553, + 689 + ], + "type": "text", + "content": "Training. Our enhanced multi-view 2D VTON network can be trained by minimizing the following latent diffusion model loss function:" + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 339, + 700, + 553, + 715 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 339, + 700, + 553, + 715 + ], + "spans": [ + { + "bbox": [ + 339, + 700, + 553, + 715 + ], + "type": "interline_equation", + "content": "\\mathcal {L} _ {\\mathrm {l d m}} = \\mathbb {E} _ {z _ {t}, \\eta , \\psi , \\epsilon , t} \\left[ \\| \\epsilon - \\epsilon_ {\\theta} (z _ {t}, t, \\eta , \\psi , \\zeta)) \\| _ {2} ^ {2} \\right], \\tag {6}", + "image_path": "7060e82bdd07dd2211be1445f3177a9dade7c07db6214dd6a3cef51337040564.jpg" + } + ] + } + ], + "index": 16 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "26392" + } + ] + } + ], + "index": 17 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 4 + }, + { + "para_blocks": [ + { + "type": "image", + "bbox": [ + 57, + 70, + 298, + 233 + ], + "blocks": [ + { + "bbox": [ + 57, + 70, + 298, + 233 + ], + "lines": [ + { + "bbox": [ + 57, + 70, + 298, + 233 + ], + "spans": [ + { + "bbox": [ + 57, + 70, + 298, + 233 + ], + "type": "image", + "image_path": "66efb9bdfdf378e1471d23c74712a4482299aa59ce3a5ebfeba941baa74bac58.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "lines": [ + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "spans": [ + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "text", + "content": "Figure 4. Illustration of the proposed Multi-view Spatial Attention. Query (Q): multi-view features " + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "inline_equation", + "content": "\\mathbf{F}^1" + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "text", + "content": "; Key (K) and Value (V): concatenation of " + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "inline_equation", + "content": "\\mathbf{F}^1" + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "text", + "content": " and garment features " + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "inline_equation", + "content": "F_{f}^{l}" + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "inline_equation", + "content": "F_{b}^{l}" + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "text", + "content": ". The attention score between viewpoints " + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "inline_equation", + "content": "i" + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "inline_equation", + "content": "j" + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "text", + "content": " is modulated by a weight " + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "inline_equation", + "content": "C_{ij}" + }, + { + "bbox": [ + 55, + 239, + 298, + 297 + ], + "type": "text", + "content": ", determined by the cosine of the angle between them." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "image_caption" + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 301, + 296, + 351 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 301, + 296, + 351 + ], + "spans": [ + { + "bbox": [ + 55, + 301, + 296, + 351 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 55, + 301, + 296, + 351 + ], + "type": "inline_equation", + "content": "\\eta = [\\mathcal{E}(g_f); \\mathcal{E}(g_b); \\mathcal{E}(\\mathbf{N_i})_{i=1}^m]" + }, + { + "bbox": [ + 55, + 301, + 296, + 351 + ], + "type": "text", + "content": " represents the input latent garment images and latent normal maps; " + }, + { + "bbox": [ + 55, + 301, + 296, + 351 + ], + "type": "inline_equation", + "content": "\\zeta = [\\mathcal{E}'(\\mathbf{A_i})_{i=1}^m]" + }, + { + "bbox": [ + 55, + 301, + 296, + 351 + ], + "type": "text", + "content": " denotes the input latent clothing-agnostic images; " + }, + { + "bbox": [ + 55, + 301, + 296, + 351 + ], + "type": "inline_equation", + "content": "\\psi = \\mathbf{Y}" + }, + { + "bbox": [ + 55, + 301, + 296, + 351 + ], + "type": "text", + "content": " is the proposed multi-view CLIP embedding." + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 361, + 138, + 376 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 361, + 138, + 376 + ], + "spans": [ + { + "bbox": [ + 55, + 361, + 138, + 376 + ], + "type": "text", + "content": "5. Experiments" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 381, + 173, + 396 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 381, + 173, + 396 + ], + "spans": [ + { + "bbox": [ + 55, + 381, + 173, + 396 + ], + "type": "text", + "content": "5.1. Experimental Setup" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 54, + 399, + 295, + 614 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 399, + 295, + 614 + ], + "spans": [ + { + "bbox": [ + 54, + 399, + 295, + 614 + ], + "type": "text", + "content": "Datasets. We conduct experiments on two public datasets: Thuman2.0 [55] and MVHumanNet [51]. Thuman2.0 comprises 526 reconstructed clothed human scans, from which we render multi-view input images. Of these samples, 426 are used for training, while the remaining 100 are set aside for testing. To further evaluate the effectiveness and robustness of our method, we also perform experiments on MVHumanNet, a large-scale dataset of multi-view human images that encompasses a diverse range of subjects, daily outfits and motion sequences. The images in MVHumanNet are captured using a multi-view system with either 48 or 24 cameras. We use 4,990 subjects from this dataset, allocating 4,790 to training and 200 for tests. For each subject, we randomly select two frames of multi-view images from its entire motion sequence. While MVHumanNet provides multi-view images directly for editing and reconstruction, we render uniformly distributed views around each human subject in Thuman2.0 to ensure consistent input." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 618, + 295, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 618, + 295, + 715 + ], + "spans": [ + { + "bbox": [ + 55, + 618, + 295, + 715 + ], + "type": "text", + "content": "Baselines. We primarily compare our method with three existing methods: DreamWaltz [23], GaussCtrl [48], and TIP-Editor [63]. DreamWaltz is a method designed for directly generating 3D human bodies based on textual descriptions, while GaussCtrl and TIP-Editor are two radiance-based editing methods. GaussCtrl is based on Stable Diffusion, using a description-like prompt to edit the scene. TIP-Editor accepts both text and image prompts. We configure it by" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 313, + 72, + 553, + 120 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 72, + 553, + 120 + ], + "spans": [ + { + "bbox": [ + 313, + 72, + 553, + 120 + ], + "type": "text", + "content": "specifying the human body as the editing region and the desired garment as the image prompt. We use ChatGPT to generate the text prompts corresponding to the clothing images." + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 313, + 123, + 555, + 278 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 123, + 555, + 278 + ], + "spans": [ + { + "bbox": [ + 313, + 123, + 555, + 278 + ], + "type": "text", + "content": "Evaluation Metrics. For quantitative evaluation, we assess garment-to-person alignment between the edited person and the reference image. Following [63], we calculate the average DINO similarity [34] between the reference image and the rendered multi-view images of the edited 3D scene. Additionally, to evaluate multi-view consistency, we compute the CLIP Directional Consistency Score as outlined in [17]. Given the large scale of experiments (repeated 3DGS reconstruction), we selected a subset of examples from the dataset for metric evaluation. Specifically, from the test sets of Thuman and MVHumanNet, we randomly sampled 10 human scans each, performing virtual try-on with 6 randomly chosen garments per human scan." + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 313, + 278, + 556, + 387 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 278, + 556, + 387 + ], + "spans": [ + { + "bbox": [ + 313, + 278, + 556, + 387 + ], + "type": "text", + "content": "We further conducted a user study involving 50 participants who rated the results of our method and three baseline methods based on two criteria: overall \"Quality\" and \"Alignment\" with the reference image. Each evaluation consisted of two questions: (1) Which method produces the highest quality of the edited 3D human? and (2) Which method achieves the most consistent alignment with the target clothing? Participants viewed the VTON results as rotating randomized video sequences." + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 389, + 556, + 484 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 389, + 556, + 484 + ], + "spans": [ + { + "bbox": [ + 313, + 389, + 556, + 484 + ], + "type": "text", + "content": "Implementation Details. During pre-processing, we crop the multi-view images to the bounding box around the person and resize them to a resolution of " + }, + { + "bbox": [ + 313, + 389, + 556, + 484 + ], + "type": "inline_equation", + "content": "768 \\times 576" + }, + { + "bbox": [ + 313, + 389, + 556, + 484 + ], + "type": "text", + "content": ". The front view and the back view of garment images are obtained from the corresponding clothed human images. After editing, we pad the images back to their original size. The data processing pipeline is the same for both Thuman2.0 and MVHumanNet datasets." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 313, + 485, + 556, + 665 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 485, + 556, + 665 + ], + "spans": [ + { + "bbox": [ + 313, + 485, + 556, + 665 + ], + "type": "text", + "content": "The Main UNet and the GarmentNet are initialized by the Stable Diffusion V1.5 [39]. The training process is divided into two stages. In the first stage, each view is trained independently, during which we establish the feature extraction capabilities of both the PoseEncoder and GarmentNet, as well as the generative capability of the Main UNet. The second stage involves multi-view training, where we randomly select " + }, + { + "bbox": [ + 313, + 485, + 556, + 665 + ], + "type": "inline_equation", + "content": "M" + }, + { + "bbox": [ + 313, + 485, + 556, + 665 + ], + "type": "text", + "content": " views for each human subject. This stage is focused on training the proposed MVAttention module to enhance the network's ability to maintain consistency across views. Due to memory constraints, we set " + }, + { + "bbox": [ + 313, + 485, + 556, + 665 + ], + "type": "inline_equation", + "content": "M = 8" + }, + { + "bbox": [ + 313, + 485, + 556, + 665 + ], + "type": "text", + "content": " for the training phase. During the testing phase, we uniformly sampled 32 views from a 360-degree rotation around the subject. The editing of these 32 views is conducted in two batches, with each batch processing " + }, + { + "bbox": [ + 313, + 485, + 556, + 665 + ], + "type": "inline_equation", + "content": "M = 16" + }, + { + "bbox": [ + 313, + 485, + 556, + 665 + ], + "type": "text", + "content": " views." + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 313, + 672, + 544, + 685 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 672, + 544, + 685 + ], + "spans": [ + { + "bbox": [ + 313, + 672, + 544, + 685 + ], + "type": "text", + "content": "5.2. Comparisons with State-of-the-Art Methods" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 313, + 689, + 555, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 689, + 555, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 689, + 555, + 715 + ], + "type": "text", + "content": "Qualitative Evaluation. Fig. 5 shows visual comparisons between our method and the baselines. DreamWaltz [23]" + } + ] + } + ], + "index": 13 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "text", + "content": "26393" + } + ] + } + ], + "index": 14 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 5 + }, + { + "para_blocks": [ + { + "type": "table", + "bbox": [ + 59, + 70, + 552, + 143 + ], + "blocks": [ + { + "bbox": [ + 59, + 70, + 552, + 143 + ], + "lines": [ + { + "bbox": [ + 59, + 70, + 552, + 143 + ], + "spans": [ + { + "bbox": [ + 59, + 70, + 552, + 143 + ], + "type": "table", + "html": "
MethodThuman2.0 [55]MVHumanNet [51]
CLIPcons↑DINOsim↑VotequalityVotealignCLIPcons↑DINOsim↑VotequalityVotealign
DreamWaltz [23]0.8870.5560.46%1.54%0.9350.4950.46%0.46%
TIP-Editor [63]0.9390.5690.92%0.62%0.9480.5122.15%1.38%
GaussCtrl [48]0.9310.5771.08%1.38%0.9380.5211.69%1.23%
Ours0.9230.63397.54%96.46%0.9330.62395.69%96.92%
", + "image_path": "16b7e1376295cee170255b3f0d8c6c37631ea1766a2dacc28c5a68023f4e5152.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "table_body" + } + ], + "index": 0 + }, + { + "bbox": [ + 70, + 147, + 538, + 159 + ], + "lines": [ + { + "bbox": [ + 70, + 147, + 538, + 159 + ], + "spans": [ + { + "bbox": [ + 70, + 147, + 538, + 159 + ], + "type": "text", + "content": "Table 1. Quantitative comparisons. " + }, + { + "bbox": [ + 70, + 147, + 538, + 159 + ], + "type": "inline_equation", + "content": "{\\mathrm{{CLIP}}}_{\\text{cons }}" + }, + { + "bbox": [ + 70, + 147, + 538, + 159 + ], + "type": "text", + "content": " denotes the CLIP Direction Consistency Score. " + }, + { + "bbox": [ + 70, + 147, + 538, + 159 + ], + "type": "inline_equation", + "content": "{\\mathrm{{DINO}}}_{\\text{sim }}" + }, + { + "bbox": [ + 70, + 147, + 538, + 159 + ], + "type": "text", + "content": " is the DINO similarity." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "text" + }, + { + "type": "image", + "bbox": [ + 82, + 161, + 531, + 485 + ], + "blocks": [ + { + "bbox": [ + 82, + 161, + 531, + 485 + ], + "lines": [ + { + "bbox": [ + 82, + 161, + 531, + 485 + ], + "spans": [ + { + "bbox": [ + 82, + 161, + 531, + 485 + ], + "type": "image", + "image_path": "6b71985790ddfa2ccbcfe297ed7bf36c95d426c4edc7ae3f9d723d3ce1a7d5c3.jpg" + } + ] + } + ], + "index": 2, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 54, + 495, + 555, + 529 + ], + "lines": [ + { + "bbox": [ + 54, + 495, + 555, + 529 + ], + "spans": [ + { + "bbox": [ + 54, + 495, + 555, + 529 + ], + "type": "text", + "content": "Figure 5. Qualitative comparison. The first two rows show the results on Thuman2.0 dataset while the last two rows show the results on MVHumaNet dataset. Our method achieves good texture preservation (highlighted by the blue boxes), while three baseline methods mostly fail." + } + ] + } + ], + "index": 3, + "angle": 0, + "type": "image_caption" + } + ], + "index": 2 + }, + { + "bbox": [ + 54, + 537, + 294, + 657 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 537, + 294, + 657 + ], + "spans": [ + { + "bbox": [ + 54, + 537, + 294, + 657 + ], + "type": "text", + "content": "regenerates 3D clothed humans from text prompts but struggles to accurately retain both body and clothing characteristics. GaussCtrl [48], lacking support for image prompts, fails to maintain detailed clothing textures. While TipEditor [63] leverages Lora [21] for personalization, it encounters difficulties in consistently mapping clothing inputs from two views into the 3D human because the personalized concept are semantic in 2D space. In contrast, our method effectively preserves intricate clothing details, such as text, stripes, and logos." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 54, + 665, + 295, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 665, + 295, + 715 + ], + "spans": [ + { + "bbox": [ + 54, + 665, + 295, + 715 + ], + "type": "text", + "content": "Quantitative Evaluation. Tab. 1 shows the results for the CLIP Directional Consistency Score and DINO similarity on Thuman2.0 and MVHumanNet datasets. Our approach surpasses other methods on " + }, + { + "bbox": [ + 54, + 665, + 295, + 715 + ], + "type": "inline_equation", + "content": "\\mathrm{DINO}_{sim}" + }, + { + "bbox": [ + 54, + 665, + 295, + 715 + ], + "type": "text", + "content": ", clearly illustrating" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 313, + 537, + 555, + 657 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 537, + 555, + 657 + ], + "spans": [ + { + "bbox": [ + 313, + 537, + 555, + 657 + ], + "type": "text", + "content": "the superiority of our method in terms of garment texture preservation. While our results on " + }, + { + "bbox": [ + 313, + 537, + 555, + 657 + ], + "type": "inline_equation", + "content": "\\mathrm{CLIP}_{\\text {cons }}" + }, + { + "bbox": [ + 313, + 537, + 555, + 657 + ], + "type": "text", + "content": " are comparable to those of other methods, it is important to note that these methods incorporate the SDS loss, which to some extent smoothes the representation of humans in 3D space. Additionally, the \"flatter\" textures of other methods could also result in artificially higher consistency scores. Furthermore, user studies have shown that our method significantly exceeds baselines in terms of edited 3D human quality and the alignment of clothing details." + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 313, + 670, + 538, + 683 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 670, + 538, + 683 + ], + "spans": [ + { + "bbox": [ + 313, + 670, + 538, + 683 + ], + "type": "text", + "content": "5.3. Visual Results using E-commerce Garment" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 313, + 689, + 555, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 689, + 555, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 689, + 555, + 715 + ], + "type": "text", + "content": "Fig. 6 showcases VTON results using garments from the MVG dataset [46], whose images are from e-commerce" + } + ] + } + ], + "index": 8 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "text", + "content": "26394" + } + ] + } + ], + "index": 9 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 6 + }, + { + "para_blocks": [ + { + "type": "image", + "bbox": [ + 58, + 71, + 294, + 270 + ], + "blocks": [ + { + "bbox": [ + 58, + 71, + 294, + 270 + ], + "lines": [ + { + "bbox": [ + 58, + 71, + 294, + 270 + ], + "spans": [ + { + "bbox": [ + 58, + 71, + 294, + 270 + ], + "type": "image", + "image_path": "7737d81fbc34085b8a538160848be10370bc9d5255819ab87e19ccf3b3907f3f.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 55, + 285, + 296, + 353 + ], + "lines": [ + { + "bbox": [ + 55, + 285, + 296, + 353 + ], + "spans": [ + { + "bbox": [ + 55, + 285, + 296, + 353 + ], + "type": "text", + "content": "Figure 6. Generalization to e-commerce garments (the MVG dataset). Our method, trained on the THuman2.0 dataset, shows strong generalizability when applied to e-commerce garments. For clarity in visualization, we display garment images on human models; however, in the actual VTON process, the garments are segmented from the models using parse maps." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "image_caption" + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 361, + 296, + 456 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 361, + 296, + 456 + ], + "spans": [ + { + "bbox": [ + 55, + 361, + 296, + 456 + ], + "type": "text", + "content": "platforms like YOOX NET-A-PORTER, Taobao, and TikTok*, and a model trained on the Thuman2.0 dataset [55]. The results demonstrate that our method effectively preserves intricate garment details and textures. For instance, it accurately maintains the stripe patterns in the first row, the cute tie in the second row, and the buttons in the third row, highlighting the robustness of our approach in handling diverse and realistic clothing items." + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 468, + 149, + 480 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 468, + 149, + 480 + ], + "spans": [ + { + "bbox": [ + 55, + 468, + 149, + 480 + ], + "type": "text", + "content": "5.4. Ablation Study" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 487, + 296, + 689 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 487, + 296, + 689 + ], + "spans": [ + { + "bbox": [ + 55, + 487, + 296, + 689 + ], + "type": "text", + "content": "We conduct an ablation study on Thuman2.0 dataset in Tab. 2 and Fig. 7 to evaluate the impact of our three proposed modules in enhancing a typical 2D VTON network with 3D-consistent generation capabilities. Starting with the 2D VTON baseline [53] using DensePose, we progressively replace DensePose with our pseudo-3D pose, incorporate multi-view CLIP embeddings, and ultimately integrate MVAttention in the final configuration. Results in Tab. 2 indicate that each module contributes to metric improvements. Fig. 7 visualizes an example of multi-view image editing. The incorporation of pseudo-3D pose substantially improves limb generation compared to the 2D VTON baseline. Comparing rows 4 and 5, prior to the integration of multi-view CLIP embedding, the model captures limited spatial information, resulting in detail loss at specific angles (columns 3, 4, and 6). Finally, the proposed MVAttention achieves a more coherent generation across views." + } + ] + } + ], + "index": 4 + }, + { + "type": "image", + "bbox": [ + 328, + 70, + 541, + 327 + ], + "blocks": [ + { + "bbox": [ + 328, + 70, + 541, + 327 + ], + "lines": [ + { + "bbox": [ + 328, + 70, + 541, + 327 + ], + "spans": [ + { + "bbox": [ + 328, + 70, + 541, + 327 + ], + "type": "image", + "image_path": "5de364b5b01530170b9e9a96febaa852e03b779b915dc80ff95da76bb7a36c53.jpg" + } + ] + } + ], + "index": 5, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 313, + 330, + 555, + 407 + ], + "lines": [ + { + "bbox": [ + 313, + 330, + 555, + 407 + ], + "spans": [ + { + "bbox": [ + 313, + 330, + 555, + 407 + ], + "type": "text", + "content": "Figure 7. Visualization of the impact of the three proposed techniques on multi-view consistent editing. The red boxes highlight the artifacts. Starting from the 2D VTON baseline, the pseudo-3D pose improves limb generation quality, multi-view CLIP embedding enhances detail across different viewing directions, and finally, MVAttention further strengthens consistency in the generated images." + } + ] + } + ], + "index": 6, + "angle": 0, + "type": "image_caption" + } + ], + "index": 5 + }, + { + "type": "table", + "bbox": [ + 317, + 417, + 550, + 491 + ], + "blocks": [ + { + "bbox": [ + 317, + 417, + 550, + 491 + ], + "lines": [ + { + "bbox": [ + 317, + 417, + 550, + 491 + ], + "spans": [ + { + "bbox": [ + 317, + 417, + 550, + 491 + ], + "type": "table", + "html": "
MethodsCLIPcons↑DINOsim↑
2D-VTON0.8920.609
+ Pseudo-3D Pose0.9100.626
+ Multi-view CLIP Embedding0.9130.631
+ Multi-view Spatial Attention0.9230.633
", + "image_path": "f705e5297b3e2f00053ee946d9e1e39d2ea67a67e62a277374d728081f02ef49.jpg" + } + ] + } + ], + "index": 7, + "angle": 0, + "type": "table_body" + } + ], + "index": 7 + }, + { + "bbox": [ + 313, + 494, + 554, + 516 + ], + "lines": [ + { + "bbox": [ + 313, + 494, + 554, + 516 + ], + "spans": [ + { + "bbox": [ + 313, + 494, + 554, + 516 + ], + "type": "text", + "content": "Table 2. Ablation studies. We ablate the impact of the three proposed techniques on Thuman2.0 dataset." + } + ] + } + ], + "index": 8, + "angle": 0, + "type": "text" + }, + { + "bbox": [ + 314, + 533, + 392, + 546 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 533, + 392, + 546 + ], + "spans": [ + { + "bbox": [ + 314, + 533, + 392, + 546 + ], + "type": "text", + "content": "6. Conclusions" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 557, + 555, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 557, + 555, + 713 + ], + "spans": [ + { + "bbox": [ + 313, + 557, + 555, + 713 + ], + "type": "text", + "content": "In this work, we proposed VTON 360, a novel 3D Virtual Try-On (VTON) method that achieves high-fidelity VTON with the ability to render clothing from arbitrary viewing directions. Our method features a novel formulation of 3D VTON as an extension of 2D VTON that ensures 3D consistent results across multiple views. To bridge the gap between 2D VTON models and 3D consistency requirements, we introduce several key innovations, including multi-view inputs, pseudo-3D pose representation, multi-view spatial attention, and multi-view CLIP embedding. Extensive experiments demonstrate the effectiveness of our approach, significantly outperforming prior 3D VTON techniques in both fidelity and any-view rendering." + } + ] + } + ], + "index": 10 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 67, + 702, + 269, + 713 + ], + "type": "page_footnote", + "angle": 0, + "lines": [ + { + "bbox": [ + 67, + 702, + 269, + 713 + ], + "spans": [ + { + "bbox": [ + 67, + 702, + 269, + 713 + ], + "type": "text", + "content": "* https://net-a-porter.com, www.taobao.com, www.douyin.com" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "26395" + } + ] + } + ], + "index": 12 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 7 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 72, + 153, + 85 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 72, + 153, + 85 + ], + "spans": [ + { + "bbox": [ + 56, + 72, + 153, + 85 + ], + "type": "text", + "content": "Acknowledgement" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 91, + 296, + 163 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 91, + 296, + 163 + ], + "spans": [ + { + "bbox": [ + 55, + 91, + 296, + 163 + ], + "type": "text", + "content": "This work is supported in part by the National Key R&D Program of China under Grant No.2024YFB3908503, in part by the National Natural Science Foundation of China under Grant NO. 62322608 and in part by the CCF-Kuaishou Large Model Explorer Fund (NO. CCF-KuaiShou 2024007)." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 175, + 115, + 188 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 175, + 115, + 188 + ], + "spans": [ + { + "bbox": [ + 56, + 175, + 115, + 188 + ], + "type": "text", + "content": "References" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 57, + 196, + 296, + 713 + ], + "type": "list", + "angle": 0, + "index": 13, + "blocks": [ + { + "bbox": [ + 61, + 196, + 295, + 217 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 61, + 196, + 295, + 217 + ], + "spans": [ + { + "bbox": [ + 61, + 196, + 295, + 217 + ], + "type": "text", + "content": "[1] Easymocap - make human motion capture easier. Github, 2021.5" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 61, + 219, + 296, + 262 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 61, + 219, + 296, + 262 + ], + "spans": [ + { + "bbox": [ + 61, + 219, + 296, + 262 + ], + "type": "text", + "content": "[2] Shuai Bai, Huiling Zhou, Zhikang Li, Chang Zhou, and Hongxia Yang. Single stage virtual try-on via deformable attention flows. In European Conference on Computer Vision, pages 409-425. Springer, 2022. 2" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 62, + 265, + 296, + 319 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 265, + 296, + 319 + ], + "spans": [ + { + "bbox": [ + 62, + 265, + 296, + 319 + ], + "type": "text", + "content": "[3] Bharat Lal Bhatnagar, Garvita Tiwari, Christian Theobalt, and Gerard Pons-Moll. Multi-garment net: Learning to dress 3d people from images. In Proceedings of the IEEE/CVF international conference on computer vision, pages 5420-5430, 2019. 2, 3" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 62, + 321, + 295, + 374 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 321, + 295, + 374 + ], + "spans": [ + { + "bbox": [ + 62, + 321, + 295, + 374 + ], + "type": "text", + "content": "[4] Robert Bridson, Ronald Fedkiw, and John Anderson. Robust treatment of collisions, contact and friction for cloth animation. In Proceedings of the 29th annual conference on Computer graphics and interactive techniques, pages 594-603, 2002. 3" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 62, + 377, + 296, + 430 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 377, + 296, + 430 + ], + "spans": [ + { + "bbox": [ + 62, + 377, + 296, + 430 + ], + "type": "text", + "content": "[5] Tim Brooks, Aleksander Holynski, and Alexei A Efros. Instructpix2pix: Learning to follow image editing instructions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18392-18402, 2023. 3" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 62, + 434, + 296, + 487 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 434, + 296, + 487 + ], + "spans": [ + { + "bbox": [ + 62, + 434, + 296, + 487 + ], + "type": "text", + "content": "[6] Haodong Chen, Yongle Huang, Haojian Huang, Xiangsheng Ge, and Dian Shao. Gaussianvton: 3d human virtual try-on via multi-stage gaussian splatting editing with image prompting. arXiv preprint arXiv:2405.07472, 2024. 2, 3, 1" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 62, + 490, + 296, + 555 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 490, + 296, + 555 + ], + "spans": [ + { + "bbox": [ + 62, + 490, + 296, + 555 + ], + "type": "text", + "content": "[7] Yiwen Chen, Zilong Chen, Chi Zhang, Feng Wang, Xiaofeng Yang, Yikai Wang, Zhongang Cai, Lei Yang, Huaping Liu, and Guosheng Lin. Gaussianeditor: Swift and controllable 3d editing with gaussian splatting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21476-21485, 2024. 3" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 62, + 557, + 296, + 612 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 557, + 296, + 612 + ], + "spans": [ + { + "bbox": [ + 62, + 557, + 296, + 612 + ], + "type": "text", + "content": "[8] Seunghwan Choi, Sunghyun Park, Minsoo Lee, and Jaegul Choo. Viton-hd: High-resolution virtual try-on via misalignment-aware normalization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14131-14140, 2021. 1, 2" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 62, + 613, + 296, + 656 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 613, + 296, + 656 + ], + "spans": [ + { + "bbox": [ + 62, + 613, + 296, + 656 + ], + "type": "text", + "content": "[9] Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for virtual try-on. arXiv preprint arXiv:2403.05139, 2024. 3, 4, 5" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 57, + 658, + 295, + 713 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 658, + 295, + 713 + ], + "spans": [ + { + "bbox": [ + 57, + 658, + 295, + 713 + ], + "type": "text", + "content": "[10] Xin Dong, Fuwei Zhao, Zhenyu Xie, Xijin Zhang, Daniel K Du, Min Zheng, Xiang Long, Xiaodan Liang, and Jianchao Yang. Dressing in the wild by watching dance videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3480-3489, 2022. 2" + } + ] + } + ], + "index": 12 + } + ], + "sub_type": "ref_text" + }, + { + "bbox": [ + 316, + 73, + 553, + 713 + ], + "type": "list", + "angle": 0, + "index": 28, + "blocks": [ + { + "bbox": [ + 316, + 73, + 553, + 127 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 73, + 553, + 127 + ], + "spans": [ + { + "bbox": [ + 316, + 73, + 553, + 127 + ], + "type": "text", + "content": "[11] Yuying Ge, Yibing Song, Ruimao Zhang, Chongjian Ge, Wei Liu, and Ping Luo. Parser-free virtual try-on via distilling appearance flows. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8485-8493, 2021. 2" + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 317, + 129, + 553, + 173 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 129, + 553, + 173 + ], + "spans": [ + { + "bbox": [ + 317, + 129, + 553, + 173 + ], + "type": "text", + "content": "[12] Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang. Taming the power of diffusion models for high-quality virtual try-on with appearance flow. arXiv preprint arXiv:2308.06101, 2023. 1, 2, 4" + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 317, + 175, + 553, + 217 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 175, + 553, + 217 + ], + "spans": [ + { + "bbox": [ + 317, + 175, + 553, + 217 + ], + "type": "text", + "content": "[13] Peng Guan, Loretta Reiss, David A Hirshberg, Alexander Weiss, and Michael J Black. Drape: Dressing any person. ACM Transactions on Graphics (ToG), 31(4):1-10, 2012. 1, 3" + } + ] + } + ], + "index": 16 + }, + { + "bbox": [ + 316, + 219, + 553, + 262 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 219, + 553, + 262 + ], + "spans": [ + { + "bbox": [ + 316, + 219, + 553, + 262 + ], + "type": "text", + "content": "[14] Riza Alp Güler, Natalia Neverova, and Iasonas Kokkinos. Densepose: Dense human pose estimation in the wild. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7297-7306, 2018. 4, 5" + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 317, + 264, + 553, + 319 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 264, + 553, + 319 + ], + "spans": [ + { + "bbox": [ + 317, + 264, + 553, + 319 + ], + "type": "text", + "content": "[15] Fabian Hahn, Bernhard Thomaszewski, Stelian Coros, Robert W Sumner, Forrester Cole, Mark Meyer, Tony DeRose, and Markus Gross. Subspace clothing simulation using adaptive bases. ACM Transactions on Graphics (TOG), 33(4):1-9, 2014. 1, 3" + } + ] + } + ], + "index": 18 + }, + { + "bbox": [ + 317, + 321, + 553, + 364 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 321, + 553, + 364 + ], + "spans": [ + { + "bbox": [ + 317, + 321, + 553, + 364 + ], + "type": "text", + "content": "[16] Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S Davis. Viton: An image-based virtual try-on network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7543-7552, 2018. 2" + } + ] + } + ], + "index": 19 + }, + { + "bbox": [ + 317, + 365, + 553, + 420 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 365, + 553, + 420 + ], + "spans": [ + { + "bbox": [ + 317, + 365, + 553, + 420 + ], + "type": "text", + "content": "[17] Ayaan Haque, Matthew Tancik, Alexei A Efros, Aleksander Holynski, and Angjoo Kanazawa. Instruct-nerf2nerf: Editing 3d scenes with instructions. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 19740–19750, 2023. 3, 6, 1" + } + ] + } + ], + "index": 20 + }, + { + "bbox": [ + 317, + 422, + 553, + 464 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 422, + 553, + 464 + ], + "spans": [ + { + "bbox": [ + 317, + 422, + 553, + 464 + ], + "type": "text", + "content": "[18] Sen He, Yi-Zhe Song, and Tao Xiang. Style-based global appearance flow for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3470-3479, 2022. 2" + } + ] + } + ], + "index": 21 + }, + { + "bbox": [ + 317, + 467, + 553, + 521 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 467, + 553, + 521 + ], + "spans": [ + { + "bbox": [ + 317, + 467, + 553, + 521 + ], + "type": "text", + "content": "[19] Zijian He, Peixin Chen, Guangrun Wang, Guanbin Li, Philip HS Torr, and Liang Lin. Wildvidfit: Video virtual try-on in the wild via image-based controlled diffusion models. In European Conference on Computer Vision, pages 123-139. Springer, 2024. 1, 2" + } + ] + } + ], + "index": 22 + }, + { + "bbox": [ + 317, + 522, + 553, + 555 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 522, + 553, + 555 + ], + "spans": [ + { + "bbox": [ + 317, + 522, + 553, + 555 + ], + "type": "text", + "content": "[20] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in Neural Information Processing Systems, 33:6840-6851, 2020. 3" + } + ] + } + ], + "index": 23 + }, + { + "bbox": [ + 317, + 556, + 553, + 600 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 556, + 553, + 600 + ], + "spans": [ + { + "bbox": [ + 317, + 556, + 553, + 600 + ], + "type": "text", + "content": "[21] Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021. 2, 3, 7" + } + ] + } + ], + "index": 24 + }, + { + "bbox": [ + 317, + 601, + 553, + 645 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 601, + 553, + 645 + ], + "spans": [ + { + "bbox": [ + 317, + 601, + 553, + 645 + ], + "type": "text", + "content": "[22] Li Hu, Xin Gao, Peng Zhang, Ke Sun, Bang Zhang, and Liefeng Bo. Animate anyone: Consistent and controllable image-to-video synthesis for character animation. arXiv preprint arXiv:2311.17117, 2023. 5" + } + ] + } + ], + "index": 25 + }, + { + "bbox": [ + 317, + 647, + 553, + 689 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 647, + 553, + 689 + ], + "spans": [ + { + "bbox": [ + 317, + 647, + 553, + 689 + ], + "type": "text", + "content": "[23] Yukun Huang, Jianan Wang, Ailing Zeng, He Cao, Xianbiao Qi, Yukai Shi, Zheng-Jun Zha, and Lei Zhang. Dreamwaltz: Make a scene with complex 3d animatable avatars. Advances in Neural Information Processing Systems, 36, 2024. 3, 6, 7" + } + ] + } + ], + "index": 26 + }, + { + "bbox": [ + 316, + 692, + 553, + 713 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 692, + 553, + 713 + ], + "spans": [ + { + "bbox": [ + 316, + 692, + 553, + 713 + ], + "type": "text", + "content": "[24] Yangyi Huang, Hongwei Yi, Yuliang Xiu, Tingting Liao, Ji-axiang Tang, Deng Cai, and Justus Thies. Tech: Text-guided" + } + ] + } + ], + "index": 27 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 318, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 757 + ], + "type": "text", + "content": "26396" + } + ] + } + ], + "index": 29 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 8 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 72, + 295, + 713 + ], + "type": "list", + "angle": 0, + "index": 14, + "blocks": [ + { + "bbox": [ + 76, + 72, + 294, + 105 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 76, + 72, + 294, + 105 + ], + "spans": [ + { + "bbox": [ + 76, + 72, + 294, + 105 + ], + "type": "text", + "content": "reconstruction of lifelike clothed humans. In 2024 International Conference on 3D Vision (3DV), pages 1531-1542. IEEE, 2024. 3" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 106, + 295, + 149 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 106, + 295, + 149 + ], + "spans": [ + { + "bbox": [ + 56, + 106, + 295, + 149 + ], + "type": "text", + "content": "[25] Bernhard Kerbl, Georgios Kopanas, Thomas Leimkuhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139-1, 2023. 3, 1" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 150, + 294, + 194 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 150, + 294, + 194 + ], + "spans": [ + { + "bbox": [ + 56, + 150, + 294, + 194 + ], + "type": "text", + "content": "[26] Jeongho Kim, Gyojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on. arXiv preprint arXiv:2312.01725, 2023. 3, 4, 5" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 57, + 195, + 294, + 216 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 195, + 294, + 216 + ], + "spans": [ + { + "bbox": [ + 57, + 195, + 294, + 216 + ], + "type": "text", + "content": "[27] Diederik P Kingma. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013. 3" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 57, + 217, + 294, + 259 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 217, + 294, + 259 + ], + "spans": [ + { + "bbox": [ + 57, + 217, + 294, + 259 + ], + "type": "text", + "content": "[28] Zorah Lahner, Daniel Cremers, and Tony Tung. Deepwrinkles: Accurate and realistic clothing modeling. In Proceedings of the European conference on computer vision (ECCV), pages 667-684, 2018. 1, 3" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 56, + 261, + 294, + 315 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 261, + 294, + 315 + ], + "spans": [ + { + "bbox": [ + 56, + 261, + 294, + 315 + ], + "type": "text", + "content": "[29] Sangyun Lee, Gyojung Gu, Sunghyun Park, Seunghwan Choi, and Jaegul Choo. High-resolution virtual try-on with misalignment and occlusion-handled conditions. In Proceedings of the European conference on computer vision (ECCV), 2022. 2" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 56, + 316, + 294, + 360 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 316, + 294, + 360 + ], + "spans": [ + { + "bbox": [ + 56, + 316, + 294, + 360 + ], + "type": "text", + "content": "[30] Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J Black. Smpl: A skinned multiperson linear model. In Seminal Graphics Papers: Pushing the Boundaries, Volume 2, pages 851-866. 2023. 3" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 56, + 361, + 294, + 415 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 361, + 294, + 415 + ], + "spans": [ + { + "bbox": [ + 56, + 361, + 294, + 415 + ], + "type": "text", + "content": "[31] Yifang Men, Yiming Mao, Yuning Jiang, Wei-Ying Ma, and Zhouhui Lian. Controllable person image synthesis with attribute-decomposed gan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5084-5093, 2020. 2" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 56, + 415, + 294, + 469 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 415, + 294, + 469 + ], + "spans": [ + { + "bbox": [ + 56, + 415, + 294, + 469 + ], + "type": "text", + "content": "[32] Aymen Mir, Thiemo Alldieck, and Gerard Pons-Moll. Learning to transfer texture from clothing images to 3d humans. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7023-7034, 2020. 2, 3" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 56, + 471, + 294, + 514 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 471, + 294, + 514 + ], + "spans": [ + { + "bbox": [ + 56, + 471, + 294, + 514 + ], + "type": "text", + "content": "[33] Davide Morelli, Alberto Baldrati, Giuseppe Cartella, Marcelia Cornia, Marco Bertini, and Rita Cucchiara. Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on. arXiv preprint arXiv:2305.13501, 2023. 1, 2" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 56, + 515, + 294, + 569 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 515, + 294, + 569 + ], + "spans": [ + { + "bbox": [ + 56, + 515, + 294, + 569 + ], + "type": "text", + "content": "[34] Maxime Oquab, Timothee Darcet, Theo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 6" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 56, + 570, + 295, + 635 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 570, + 295, + 635 + ], + "spans": [ + { + "bbox": [ + 56, + 570, + 295, + 635 + ], + "type": "text", + "content": "[35] Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed AA Osman, Dimitrios Tzionas, and Michael J Black. Expressive body capture: 3d hands, face, and body from a single image. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10975-10985, 2019. 5" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 56, + 636, + 295, + 679 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 636, + 295, + 679 + ], + "spans": [ + { + "bbox": [ + 56, + 636, + 295, + 679 + ], + "type": "text", + "content": "[36] Gerard Pons-Moll, Sergi Pujades, Sonny Hu, and Michael J Black. Clothcap: Seamless 4d clothing capture and retargeting. ACM Transactions on Graphics (ToG), 36(4):1-15, 2017. 3" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 56, + 680, + 295, + 713 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 680, + 295, + 713 + ], + "spans": [ + { + "bbox": [ + 56, + 680, + 295, + 713 + ], + "type": "text", + "content": "[37] Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Mildenhall. Dreamfusion: Text-to-3d using 2d diffusion. arXiv preprint arXiv:2209.14988, 2022. 2, 3" + } + ] + } + ], + "index": 13 + } + ], + "sub_type": "ref_text" + }, + { + "bbox": [ + 316, + 72, + 555, + 713 + ], + "type": "list", + "angle": 0, + "index": 29, + "blocks": [ + { + "bbox": [ + 316, + 72, + 554, + 127 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 72, + 554, + 127 + ], + "spans": [ + { + "bbox": [ + 316, + 72, + 554, + 127 + ], + "type": "text", + "content": "[38] Yurui Ren, Xiaoqing Fan, Ge Li, Shan Liu, and Thomas H Li. Neural texture extraction and distribution for controllable person image synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13535-13544, 2022. 2" + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 316, + 129, + 555, + 183 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 129, + 555, + 183 + ], + "spans": [ + { + "bbox": [ + 316, + 129, + 555, + 183 + ], + "type": "text", + "content": "[39] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10684-10695, 2022. 1, 3, 4, 6" + } + ] + } + ], + "index": 16 + }, + { + "bbox": [ + 316, + 184, + 554, + 227 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 184, + 554, + 227 + ], + "spans": [ + { + "bbox": [ + 316, + 184, + 554, + 227 + ], + "type": "text", + "content": "[40] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. arXiv preprint arXiv:2208.12242, 2022. 3" + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 316, + 228, + 555, + 282 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 228, + 555, + 282 + ], + "spans": [ + { + "bbox": [ + 316, + 228, + 555, + 282 + ], + "type": "text", + "content": "[41] Igor Santesteban, Nils Thuerey, Miguel A Otaduy, and Dan Casas. Self-supervised collision handling via generative 3d garment models for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11763-11773, 2021. 2" + } + ] + } + ], + "index": 18 + }, + { + "bbox": [ + 316, + 283, + 554, + 326 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 283, + 554, + 326 + ], + "spans": [ + { + "bbox": [ + 316, + 283, + 554, + 326 + ], + "type": "text", + "content": "[42] Igor Santesteban, Miguel Otaduy, Nils Thuerey, and Dan Casas. Ulnef: Untangled layered neural fields for mix-and-match virtual try-on. Advances in Neural Information Processing Systems, 35:12110-12125, 2022. 2" + } + ] + } + ], + "index": 19 + }, + { + "bbox": [ + 316, + 327, + 554, + 380 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 327, + 554, + 380 + ], + "spans": [ + { + "bbox": [ + 316, + 327, + 554, + 380 + ], + "type": "text", + "content": "[43] Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International Conference on Machine Learning, pages 2256-2265. PMLR, 2015. 3" + } + ] + } + ], + "index": 20 + }, + { + "bbox": [ + 316, + 381, + 554, + 415 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 381, + 554, + 415 + ], + "spans": [ + { + "bbox": [ + 316, + 381, + 554, + 415 + ], + "type": "text", + "content": "[44] Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020.3" + } + ] + } + ], + "index": 21 + }, + { + "bbox": [ + 316, + 415, + 554, + 448 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 415, + 554, + 448 + ], + "spans": [ + { + "bbox": [ + 316, + 415, + 554, + 448 + ], + "type": "text", + "content": "[45] Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. Advances in Neural Information Processing Systems, 32, 2019. 3" + } + ] + } + ], + "index": 22 + }, + { + "bbox": [ + 316, + 449, + 554, + 481 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 449, + 554, + 481 + ], + "spans": [ + { + "bbox": [ + 316, + 449, + 554, + 481 + ], + "type": "text", + "content": "[46] Haoyu Wang, Zhilu Zhang, Donglin Di, Shiliang Zhang, and Wangmeng Zuo. Mv-vton: Multi-view virtual try-on with diffusion models. arXiv preprint arXiv:2404.17364, 2024. 7" + } + ] + } + ], + "index": 23 + }, + { + "bbox": [ + 316, + 482, + 554, + 535 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 482, + 554, + 535 + ], + "spans": [ + { + "bbox": [ + 316, + 482, + 554, + 535 + ], + "type": "text", + "content": "[47] Junjie Wang, Jiemin Fang, Xiaopeng Zhang, Lingxi Xie, and Qi Tian. Gaussianeditor: Editing 3d gaussians delicately with text instructions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20902-20911, 2024. 3" + } + ] + } + ], + "index": 24 + }, + { + "bbox": [ + 316, + 536, + 554, + 581 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 536, + 554, + 581 + ], + "spans": [ + { + "bbox": [ + 316, + 536, + 554, + 581 + ], + "type": "text", + "content": "[48] Jing Wu, Jia-Wang Bian, Xinghui Li, Guangrun Wang, Ian Reid, Philip Torr, and Victor Adrian Prisacariu. Gaussctrl: multi-view consistent text-driven 3d gaussian splattering editing. arXiv preprint arXiv:2403.08733, 2024. 3, 6, 7" + } + ] + } + ], + "index": 25 + }, + { + "bbox": [ + 316, + 582, + 554, + 646 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 582, + 554, + 646 + ], + "spans": [ + { + "bbox": [ + 316, + 582, + 554, + 646 + ], + "type": "text", + "content": "[49] Jay Zhangjie Wu, Yixiao Ge, Xintao Wang, Stan Weixian Lei, Yuchao Gu, Yufei Shi, Wynne Hsu, Ying Shan, Xiaohu Qie, and Mike Zheng Shou. Tune-a-video: One-shot tuning of image diffusion models for text-to-video generation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 7623-7633, 2023. 5" + } + ] + } + ], + "index": 26 + }, + { + "bbox": [ + 316, + 647, + 554, + 689 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 647, + 554, + 689 + ], + "spans": [ + { + "bbox": [ + 316, + 647, + 554, + 689 + ], + "type": "text", + "content": "[50] Zhenyu Xie, Haoye Dong, Yufei Gao, Zehua Ma, and Xiaodan Liang. Dreamvton: Customizing 3d virtual try-on with personalized diffusion models. arXiv preprint arXiv:2407.16511, 2024. 2, 3" + } + ] + } + ], + "index": 27 + }, + { + "bbox": [ + 316, + 691, + 554, + 713 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 691, + 554, + 713 + ], + "spans": [ + { + "bbox": [ + 316, + 691, + 554, + 713 + ], + "type": "text", + "content": "[51] Zhangyang Xiong, Chenghong Li, Kenkun Liu, Hongjie Liao, Jianqiao Hu, Junyi Zhu, Shuliang Ning, Lingteng Qiu," + } + ] + } + ], + "index": 28 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "26397" + } + ] + } + ], + "index": 30 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 9 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 72, + 295, + 712 + ], + "type": "list", + "angle": 0, + "index": 13, + "blocks": [ + { + "bbox": [ + 76, + 72, + 295, + 127 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 76, + 72, + 295, + 127 + ], + "spans": [ + { + "bbox": [ + 76, + 72, + 295, + 127 + ], + "type": "text", + "content": "Chongjie Wang, Shijie Wang, et al. Mvhumannet: A large-scale dataset of multi-view daily dressing human captures. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19801-19811, 2024. 2, 6, 7" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 128, + 295, + 172 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 128, + 295, + 172 + ], + "spans": [ + { + "bbox": [ + 56, + 128, + 295, + 172 + ], + "type": "text", + "content": "[52] Dejia Xu, Weili Nie, Chao Liu, Sifei Liu, Jan Kautz, Zhangyang Wang, and Arash Vahdat. Camco: Camera-controllable 3d-consistent image-to-video generation. arXiv preprint arXiv:2406.02509, 2024. 5" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 173, + 294, + 215 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 173, + 294, + 215 + ], + "spans": [ + { + "bbox": [ + 56, + 173, + 294, + 215 + ], + "type": "text", + "content": "[53] Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen. Oot-diffusion: Outfitting fusion based latent diffusion for controllable virtual try-on. arXiv preprint arXiv:2403.01779, 2024. 3, 4, 8" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 56, + 217, + 295, + 259 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 217, + 295, + 259 + ], + "spans": [ + { + "bbox": [ + 56, + 217, + 295, + 259 + ], + "type": "text", + "content": "[54] Han Yang, Xinrui Yu, and Ziwei Liu. Full-range virtual try-on with recurrent tri-level transform. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3460-3469, 2022. 2" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 56, + 261, + 294, + 325 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 261, + 294, + 325 + ], + "spans": [ + { + "bbox": [ + 56, + 261, + 294, + 325 + ], + "type": "text", + "content": "[55] Tao Yu, Zerong Zheng, Kaiwen Guo, Pengpeng Liu, Qionghai Dai, and Yebin Liu. Function4d: Real-time human volumetric capture from very sparse consumer rgbd sensors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5746-5756, 2021. 2, 6, 7, 8" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 56, + 327, + 294, + 370 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 327, + 294, + 370 + ], + "spans": [ + { + "bbox": [ + 56, + 327, + 294, + 370 + ], + "type": "text", + "content": "[56] Jinsong Zhang, Kun Li, Yu-Kun Lai, and Jingyu Yang. Pise: Person image synthesis and editing with decoupled gan. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7982-7990, 2021. 2" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 56, + 372, + 294, + 403 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 372, + 294, + 403 + ], + "spans": [ + { + "bbox": [ + 56, + 372, + 294, + 403 + ], + "type": "text", + "content": "[57] Lvmin Zhang and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. arXiv preprint arXiv:2302.05543, 2023. 3" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 56, + 404, + 294, + 448 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 404, + 294, + 448 + ], + "spans": [ + { + "bbox": [ + 56, + 404, + 294, + 448 + ], + "type": "text", + "content": "[58] Yabo Zhang, Yuxiang Wei, Dongsheng Jiang, Xiaopeng Zhang, Wangmeng Zuo, and Qi Tian. Controlvideo: Training-free controllable text-to-video generation. arXiv preprint arXiv:2305.13077, 2023. 5" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 56, + 449, + 294, + 513 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 449, + 294, + 513 + ], + "spans": [ + { + "bbox": [ + 56, + 449, + 294, + 513 + ], + "type": "text", + "content": "[59] Fuwei Zhao, Zhenyu Xie, Michael Kampffmeyer, Haoye Dong, Songfang Han, Tianxiang Zheng, Tao Zhang, and Xiaodan Liang. M3d-vton: A monocular-to-3d virtual try-on network. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 13239-13249, 2021. 3" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 56, + 514, + 294, + 579 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 514, + 294, + 579 + ], + "spans": [ + { + "bbox": [ + 56, + 514, + 294, + 579 + ], + "type": "text", + "content": "[60] Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Sahara, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two unets. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4606-4615, 2023. 2" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 56, + 581, + 294, + 624 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 581, + 294, + 624 + ], + "spans": [ + { + "bbox": [ + 56, + 581, + 294, + 624 + ], + "type": "text", + "content": "[61] Jingyu Zhuang, Chen Wang, Liang Lin, Lingjie Liu, and Guanbin Li. Dreameditor: Text-driven 3d scene editing with neural fields. In SIGGRAPH Asia 2023 Conference Papers, pages 1-10, 2023. 3" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 56, + 625, + 294, + 667 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 625, + 294, + 667 + ], + "spans": [ + { + "bbox": [ + 56, + 625, + 294, + 667 + ], + "type": "text", + "content": "[62] Jingyu Zhuang, Di Kang, Linchao Bao, Liang Lin, and Guanbin Li. Dagensm: Disentangled avatar generation with gs-enhanced mesh. arXiv preprint arXiv:2411.15205, 2024.3" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 56, + 669, + 294, + 712 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 669, + 294, + 712 + ], + "spans": [ + { + "bbox": [ + 56, + 669, + 294, + 712 + ], + "type": "text", + "content": "[63] Jingyu Zhuang, Di Kang, Yan-Pei Cao, Guanbin Li, Liang Lin, and Ying Shan. Tip-editor: An accurate 3d editor following both text-prompts and image-prompts. arXiv preprint arXiv:2401.14828, 2024. 3, 6, 7, 1" + } + ] + } + ], + "index": 12 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "text", + "content": "26398" + } + ] + } + ], + "index": 14 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 10 + } + ], + "_backend": "vlm", + "_version_name": "2.6.4" +} \ No newline at end of file diff --git a/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/00462254-bb4b-4c24-9d0a-9a8f2054757a_content_list.json b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/00462254-bb4b-4c24-9d0a-9a8f2054757a_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..ef80ed2fcc54819b52691f87fc38a7b40f5b2cef --- /dev/null +++ b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/00462254-bb4b-4c24-9d0a-9a8f2054757a_content_list.json @@ -0,0 +1,1251 @@ +[ + { + "type": "text", + "text": "VTON-HandFit: Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding", + "text_level": 1, + "bbox": [ + 124, + 128, + 875, + 176 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Yujie Liang $^{1,2*}$ , Xiaobin Hu $^{3*}$ , Boyuan Jiang $^{3}$ , Donghao Luo $^{3\\dagger}$ , Xu Peng $^{3}$ , Kai Wu $^{3}$ , Chengming Xu $^{3}$ , Wenhui Han $^{3}$ , Taisong Jin $^{1,2\\dagger}$ , Chengjie Wang $^{3}$ , Rongrong Ji $^{1,2}$ $^{1}$ Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, China. \n $^{2}$ School of Informatics, Xiamen University, China. $^{3}$ Tencent Youtu Lab.", + "bbox": [ + 179, + 202, + 820, + 292 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "liangyujie@stu.xmu.edu.cn,{xiaobinhu,byronjiang,michaelluo,ppxupeng, chengmingxu,jasoncjwang}@tencent.com,wukaiwork,outlook.com, whhan18@gmail.com,{jintaisong,rrji}@xmu.edu.cn", + "bbox": [ + 197, + 294, + 797, + 345 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract", + "text_level": 1, + "bbox": [ + 246, + 378, + 326, + 395 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Although diffusion-based image virtual try-on has made considerable progress, emerging approaches still struggle to effectively address the issue of hand occlusion (i.e., clothing regions occluded by the hand part), leading to a notable degradation of the try-on performance. To tackle this issue widely existing in real-world scenarios, we propose VTON-HandFit, leveraging the power of hand priors to reconstruct the appearance and structure for hand occlusion cases. Firstly, we tailor a Hand-Pose Aggregation Net using the ControlNet-based structure explicitly and adaptively encoding the global hand and pose priors. Besides, to fully exploit the hand-related structure and appearance information, we propose Hand-feature Disentanglement Embedding module to disentangle the hand priors into the hand structure-parametric and visual-appearance features, and customize a masked cross attention for further decoupled feature embedding. Lastly, we customize a hand-canny constraint loss to better learn the structure edge knowledge from the hand template of model image. VTON-HandFit outperforms the baselines in qualitative and quantitative evaluations on the public dataset and our self-collected hand-occlusion Handfit-3K dataset particularly for the arbitrary hand pose occlusion cases in real-world scenarios. Our project page is at: https://vton-handfit.github.io.", + "bbox": [ + 88, + 411, + 485, + 773 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1. Introduction", + "text_level": 1, + "bbox": [ + 89, + 799, + 223, + 814 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Image-based virtual try-on (VTON) [26] is a widely adopted and highly promising image synthesis technology in the e-commerce industry. Its primary goal is to enhance", + "bbox": [ + 89, + 824, + 483, + 869 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "the shopping experience for consumers and minimize advertising expenses for clothing merchants. The VTON task involves generating an image of a human model wearing a specific garment. Over the past few years, countless researchers [9-11, 50] have dedicated significant efforts to achieve more realistic and precise virtual try-on results.", + "bbox": [ + 511, + 380, + 906, + 470 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Most recent studies on virtual try-on predominantly utilize generative adversarial networks (GANs) [13] or latent diffusion models (LDMs) [39] for image synthesis. Traditional GAN-based approaches [15-17] often struggle with accurately generating garment folds, realistic lighting and shadows, and lifelike human body representations. Consequently, more recent research has shifted towards LDM-based methods [61], which effectively improve the authenticity of clothed images. However, these researches focus on generating more realistic and natural try-on images while keeping the detailed features of garments. The hands of try-on models are usually strictly on the sides and garments are not occluded by the hands. Such a strict constraint hinders the virtual try-on paving the way toward real-world applications. For instance, as a highly flexible editing task with a wide range of degrees of freedom, digital humans usually meet the occasions that the garments are occluded by the hands, which severely deteriorates the performance of virtual try-on.", + "bbox": [ + 511, + 474, + 908, + 762 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Motivated by the above challenge, basically, there mainly exist two solutions to tackle this arbitrary hand problem: 1). perfect hand parsing model to accurately segment the tiny area among fingers. 2). involving the hand-reconstruction into the try-on processing. As shown in Fig. 1, CatVTON[8] use the parsing model to keep the hand part, leading to unavoidably the remaining artifacts and backgrounds from the model image. The existing parsing models fail to accurately segment the small", + "bbox": [ + 511, + 763, + 910, + 902 + ], + "page_idx": 0 + }, + { + "type": "header", + "text": "CVF", + "bbox": [ + 106, + 2, + 181, + 42 + ], + "page_idx": 0 + }, + { + "type": "header", + "text": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore.", + "bbox": [ + 236, + 0, + 810, + 46 + ], + "page_idx": 0 + }, + { + "type": "page_footnote", + "text": "* equal contribution, † corresponding author", + "bbox": [ + 93, + 875, + 333, + 888 + ], + "page_idx": 0 + }, + { + "type": "page_number", + "text": "22616", + "bbox": [ + 478, + 945, + 519, + 955 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/ffb33d5ded6172475685713deedb9529d89316456ef43ff78a83006a4782f928.jpg", + "image_caption": [ + "Figure 1. Comparison of different models on virtual try-on with hand occlusion. The leftmost three images are reference images, showing the target clothing, the model, and a close-up of the model's hand. Our model excels in preserving hand details and achieving realistic clothing transfer, as highlighted in the red boxes. CatVTON utilizes a parsing model to retain the hand segment, resulting in the inevitable persistence of residual artifacts and backgrounds from the model image. IDM-VTON* indicates post-processing using HandRefiner, which fails to address missing hands and mistakenly alters correctly depicted ones." + ], + "image_footnote": [], + "bbox": [ + 109, + 88, + 883, + 404 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "and fragmented finger areas, thus we embrace the latter strategy to involve the hand-reconstruction by disentangling the hand into the appearance and structure knowledge processing in the try-on stage. Then, we propose a novel diffusion-based virtual try-on method, termed as VTON-HandFit, to generate realistic and natural images for arbitrary hand position or hand-occlusion cases with the aid of hand 3D structure-parametric and visual-appearance priors. To embed the pose- and hand-related knowledge, we build a Hand-Pose Net to exploit the body and hand semantic and structure information by getting the body (e.g., DWPose, DensPose) and 3D hand HaMeR-based [61] reconstruction information. Secondly, considering the over-small character of hand that is easily ignored after down-sampling operation, we use the DINO [55] to dig the local appearance and texture knowledge of hand as the hand-visual embedding features. For the structure of hand, we adopt these structure-parametric priors (e.g., 3D MANO-related parameters, handtype and hand-box coordinates) as hand-structure embedding features. After getting structure-parametric and visual-appearance features, we aim to enhance the model's ability to respectively learn hand-related patterns. To achieve this, we inject the hand-structure embedding into the Hand-Pose Net with a primary focus on enhancing the latent features related to hand structure, and the hand-visual embedding into the mainstream U-Net specifically aiming to enhance texture-related features, respec", + "bbox": [ + 91, + 489, + 483, + 896 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "tively. The disentanglement learning of structure and visual appearance enhances the network's ability to efficiently and respectively acquire hand-related knowledge. To prevent the interference of hand-related priors with garment features, we impose a constraint that restricts the hand-related features to operate solely within the hand mask area of the cross-attention mechanism. Finally, we tailor a handcanny constraint loss to more effectively acquire the structural edge knowledge from the hand template of the model image. Our contributions are summarized as follows:", + "bbox": [ + 509, + 489, + 906, + 641 + ], + "page_idx": 1 + }, + { + "type": "list", + "sub_type": "text", + "list_items": [ + "- We propose VTON-HandFit, a novel virtual try-on architecture considering to solve the Hand pose occlusion problem popularly existing in real-world scenarios.", + "- We propose a Hand-Pose Aggregation Net to encode the hand and pose structure priors and also adaptively to adjust the weight of each component. Meanwhile, hand-canny constraint loss is designed to learn the hand structural-edge knowledge of model images.", + "- We propose a novel hand-feature disentanglement embedding module disentangling the hand knowledge into the structure-parametric priors (e.g., MANO parameters) and visual-appearance priors (e.g., color). A mask cross-attention mechanism is proposed to constrain the hand-related features only working on the hand mask area to avoid the garment feature disturbance.", + "- We self-collect Handfit-3K test dataset from online retail" + ], + "bbox": [ + 506, + 647, + 908, + 898 + ], + "page_idx": 1 + }, + { + "type": "page_number", + "text": "22617", + "bbox": [ + 478, + 944, + 519, + 957 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/74037f3537d914e6a0e9804d1cc30bb05894de5764f3a14727b1395c8c65a240.jpg", + "image_caption": [ + "Figure 2. An overview of our VTON-HandFit Network. The network consists of two main components: Hand-feature Disentanglement Embedding and Hand-Pose Aggregation Net. The Hand-feature Disentanglement Embedding module uses the HaMer model to extract hand priors, including hand type $T_{h}$ , 3D vertices $V_{h}$ , spatial joint locations $J_{2d}$ , and joint rotation matrices $\\theta_{h}$ . These features are processed by the Hand-Struct processor to derive structural features $c_{struc}$ . Simultaneously, hand images cropped using bounding boxes are processed by DINOv2 and the Hand-Appear processor to obtain visual features $c_{appear}$ . The structural and visual features are integrated using mask cross attention. The Hand-Pose Aggregation Net module controls body and hand poses by aggregating DWpose, Densepose, and hand depth maps." + ], + "image_footnote": [], + "bbox": [ + 143, + 89, + 851, + 448 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "sites to evaluate hand occlusion performance. Extensive qualitative and quantitative evaluations demonstrate our superiority over state-of-the-art virtual try-on models especially for hand pose occlusion cases, paving the road towards more complicated virtual try-on in real-world.", + "bbox": [ + 96, + 566, + 483, + 643 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2. Related Works", + "text_level": 1, + "bbox": [ + 89, + 661, + 240, + 678 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Image-based virtual try-on. In the context of image-based virtual try-on, the objective is to generate a realistic representation of a target person wearing a specific garment, given a pair of corresponding images. A series of studies [26, 30, 47, 51] have been conducted using Generative Adversarial Networks (GANs) [13]. These studies typically involve the initial deformation of the garment to match the body shape of the person, followed by the utilization of a generator to superimpose the deformed garment onto the person image. Despite efforts made to minimize the discrepancy between the warped garment and the person [12, 21, 26], these approaches often lack the ability to generalize well to diverse person images, particularly those with complex backgrounds or intricate poses. There exists", + "bbox": [ + 89, + 688, + 483, + 900 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "GAN-based try-on method [26] considers occluded problems and designs the warping operation to avoid the garment warping misalignment, which is totally different from diffusion-based algorithms as an inpainting problem.", + "bbox": [ + 511, + 566, + 906, + 627 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The remarkable generative capabilities of diffusion models have sparked interest in incorporating them into fashion synthesis, particularly in tasks like visual try-on [2, 4, 5, 22, 27, 43, 59]. TryOnDiffusion [61] employs ParallelUNets to preserve garment details and warp the garment for the try-on process. Subsequent studies have treated virtual try-on as an exemplar-based image inpainting problem [49]. These studies involved fine-tuning the inpainting diffusion models using virtual try-on datasets [24, 32] to generate virtual try-on images of superior quality. For instance, such as LADI-VTON [32] and DCI-VTON [14] have emerged, treating clothing as pseudo-words or employing warping networks to integrate garments into pre-trained diffusion models. IDM-VTON [7] incorporates sophisticated attention modules to encode the high-level semantics of the garment and extract low-level features, thereby preserving fine-grained details. While these diffusion methods address issues related to the nature and realism of synthesized images,", + "bbox": [ + 511, + 628, + 908, + 902 + ], + "page_idx": 2 + }, + { + "type": "page_number", + "text": "22618", + "bbox": [ + 478, + 945, + 519, + 955 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "they struggle to meet hard occasions of the hand-based occlusion which very widely exists in the real-world virtual try-on scenarios.", + "bbox": [ + 89, + 90, + 480, + 135 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Conditional control of diffusion models. In recent years, latent diffusion models [39] have demonstrated significant success in tasks such as text-to-image generation [19, 25, 37, 40] and image-to-image generation [23, 41]. To achieve more controllable generated results, several studies have introduced conditional control to diffusion models [53, 60]. ControlNet [57] adds spatial conditioning controls (e.g.,anny edges, Hough lines, user scribbles, human key points) to pretrained diffusion models via zero convolutions manner. T2I-Adapter [33] aligns the internal knowledge within T2I models with external control signals, allowing it to be trained based on different conditions. This approach enables the generation of results with rich control and editing effects, particularly in terms of color and structure. In our paper, we focus on the hard occasions of hand-based occlusion in image-based VTON task, and tailor a Hand-Pose Aggregation Net to control the body pose and hand structure. Furthermore, the disengagement knowledge of hand structure and appearance embedding is studied and then these embeddings are respectively injected into the attention mechanism, acting as a mask-based regional control.", + "bbox": [ + 91, + 137, + 482, + 452 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Reliable human hand generation. Previous studies [42] have acknowledged the problem of misshaped hands that are generated by existing text-to-image diffusion models. Controlling the hand is considerably more challenging compared to controlling the body pose, primarily due to the relative size of the main subject. Besides, diffusion models exhibit significantly improved capability in generating plausible whole-body poses. This difficulty arises from the intricate nature of learning the physical structure and pose of hands from training images, which encompasses extensive deformations and occlusions. Handrefiner [29] and RHanDS [44] are post-processing solutions adopting a hand mesh reconstruction model to rectify malformed hands. Hand-to-Diffusion [36] uses a two-stage approach, first generating hand effects and then producing the entire image. Hand1000 [56] employs a three-stage training process that integrates hand gesture recognition and optimized text embeddings. VTON-HandFit involves the hand corrections into the generation process not as a post-processing and disentangles the hand priors into the structure and appearance feature embedding for virtual try-on arbitrary hand pose in the real-world.", + "bbox": [ + 91, + 454, + 482, + 785 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3. Methodology", + "text_level": 1, + "bbox": [ + 89, + 799, + 225, + 816 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Data collection of Handfit-3K. Considering that the hand of the current main virtual try-on dataset is separate from the body, there rarely exists a great amount of virtual try-on data for arbitrary hand pose and hard hand-occlusion cases. To meet this urgent need in real-world, we aim to", + "bbox": [ + 89, + 824, + 480, + 900 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "construct a dataset of complex hand poses that occlude the garment. Specifically, we collect an additional dataset of 26,000 images sourced from online retail sites. These images underwent processing using a parsing model and a hand bounding box via HaMeR model [35] to identify instances of occlusion. We devote significant manual effort to inspecting the quality of the data, removing any erroneous or flawed entries, thereby ensuring the high quality of the virtual try-on dataset. As a result, we obtained a specialized hand occlusion test set comprising 3,620 images, termed as Handfit-3K. This dataset facilitates a targeted evaluation of our method's ability to effectively handle occlusions.", + "bbox": [ + 511, + 90, + 903, + 272 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.1. Preliminary", + "text_level": 1, + "bbox": [ + 511, + 281, + 643, + 297 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Latent Diffusion Models. Our VTON-HandFit is an extension of stable diffusion model, which is a probabilistic model specifically designed to estimate a data distribution $\\mathfrak{p}(\\mathbf{x})$ by iteratively reducing noise in a normally distributed variable. Given an input image $x_0$ , the noise estimation process is:", + "bbox": [ + 511, + 303, + 903, + 393 + ], + "page_idx": 3 + }, + { + "type": "equation", + "text": "\n$$\n\\mathcal {L} _ {\\text {n o i s e}} = \\mathbb {E} _ {z \\sim \\mathcal {E} (x), C, \\epsilon \\sim \\mathcal {N} (0, 1), t} \\left[ | | \\epsilon - \\epsilon_ {\\theta} (z _ {t}, t, C) | | _ {2} ^ {2} \\right], \\tag {1}\n$$\n", + "text_format": "latex", + "bbox": [ + 521, + 402, + 903, + 422 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where a Variational AutoEncoder $\\mathcal{E}$ compress input image $x$ into low-dimension latent space $z$ . A conditional U-Net denoiser $\\epsilon_{\\theta}$ aims to estimate noise $\\theta$ in latent space with the aid of timestep $t$ , $t$ -th noisy latent representation $z_{t}$ , and other text-prompt conditions $C$ extracted by text-encoder.", + "bbox": [ + 511, + 430, + 903, + 506 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.2.VTON-HandFit", + "text_level": 1, + "bbox": [ + 511, + 513, + 671, + 529 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Overview. An overview of our VTON-HandFit is illustrated in Fig. 2. Given a target person image $x \\in \\mathbb{R}^{H \\times W \\times 3}$ and an input garment image $g \\in \\mathbb{R}^{H \\times W \\times 3}$ , VTON-HandFit aims to generate a realistic outfitted image $x_{g} \\in \\mathbb{R}^{H \\times W \\times 3}$ . The garment-agnostic person representation $x_{m}$ is masked to erase the garment information by the combination of HumanParsing and HumanPose. Thus, this virtual try-on task is transformed into an example-based inpainting problem to recover the missing mask part. For the input of TryonNet, we simply concatenate the noise image $(z_{t})$ and latent garment-agnostic features $\\mathcal{E}(x_{m})$ to avoid the other pose-relevant information disturbances. In Hand-Pose Aggregation Net, we also aggregate the pose-related knowledge including the DWPose, DensePose and 3D reconstruction depth map derived by the HaMeR mesh model. These pose-related priors are detached with TryonNet to explicitly be learned in Hand-Pose Net. DensePose provides more bodysemantic information, DWPose contains abundant skeleton structure, and 3D-hand-derived depth maps demonstrate the pixel-level global-aligned hand position and structure knowledge. These modules complement each other by mutually providing information, and working together to constrain the pose and hand generation. In hand-feature disentanglement embedding, we use the HaMeR model for 3D", + "bbox": [ + 511, + 537, + 903, + 900 + ], + "page_idx": 3 + }, + { + "type": "page_number", + "text": "22619", + "bbox": [ + 478, + 944, + 517, + 955 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "hand reconstruction and get the structure-parametric priors. These features are processed by the Hand-Struct processor to derive structural features $c_{struc}$ . Simultaneously, cropped hand images are processed by DINOv2 and the Hand-Appear processor to obtain visual features $c_{appear}$ . Besides, GarmentNet aims to extract the garment features into the TryonNet mainly by: 1). encoding the garment image into the latent space $\\mathcal{E}_g$ , 2). using the CLIP to get the image token features for injection.", + "bbox": [ + 89, + 90, + 480, + 226 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Hand-Pose Aggregation Net. To aggregate the hand and pose-relevant priors learning, we tailor a Hand-Pose Net to constrain the semantic and skeleton pose, and global pixel-level hand knowledge. Specifically, dwpose, densepose, and render blocks are designed based on the same zero-multilayer structure with different weights to respectively extract the corresponding features. Zero-multilayer structure consists of seven stacking of conv and SiLU activate function with the zero convolution layer at the last layer.", + "bbox": [ + 89, + 227, + 480, + 363 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\n\\mathcal {F} _ {h p} = \\mathcal {F} _ {d w} + \\mathcal {F} _ {d p} + \\mathcal {F} _ {r h} \\times w _ {\\text {h a n d}}, \\tag {2}\n$$\n", + "text_format": "latex", + "bbox": [ + 166, + 375, + 482, + 391 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $\\mathcal{F}_{dw}$ , $\\mathcal{F}_{dp}$ , $\\mathcal{F}_{rh}$ are the DWPose feature, Densepose feature, and hand feature after the Zero-multilayer structure, and then the summation is implemented on the pixel-level with the adjustable weight $w_{\\mathrm{hand}}$ of hand feature $\\mathcal{F}_{rh}$ . To mitigate the impact of detrimental noise on the hidden states of the neural network layers in the trainable Hand-Pose Net during the initial stages of training, we also adopt the zero convolution learning paradigm of ControlNet.", + "bbox": [ + 89, + 401, + 482, + 521 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Hand structural embedding. The HaMeR [35] model is utilized to reconstruct various MANO parameters and other hand-related information from 2D images, including hand mask $M_h \\in \\mathbb{R}^{768 \\times 1024}$ , hand types $T_h \\in \\mathbb{R}^{n \\times 1}$ , hand 3D vertices $V_h \\in \\mathbb{R}^{n \\times 778 \\times 3}$ , spatial joint locations $J_{2d} \\in \\mathbb{R}^{n \\times 21 \\times 2}$ , and joint rotation matrices $\\theta \\in \\mathbb{R}^{n \\times 16 \\times 3 \\times 3}$ where $n$ represents the number of hands. To accommodate variable numbers of hands as input, we pad the hand parameters to a fixed length $n = 2$ by adding padding at the end. For hand type $T_h$ , we use 0 to denote left hands, 1 for right hands, and -1 as a filler for images without hands. Similar padding is applied to other parameters to ensure alignment and correspondence across all dimensions. We apply Basis Point Set (BPS) [38] for dimensionality reduction of hand vertices and encode joint rotations as 6D vectors. And then, we import theses information as well as joint rotations and hand types into the hand-structure processor. The structure of hand-structure processor consists of two parallel blocks: three stacked layers of linear and ReLU activation $(L_r)$ and two stacked layers of linear and layer normalization $(L_n)$ . These streamlined hand vertices and joint locations are then fed into $L_r$ . Concurrently, joint rotations and hand types are transformed via $L_n$ . The outputs from these processes are aggregated to form the Hand Structural Embedding $c_{struc}$ , encapsulating essential geometric and 3D information for", + "bbox": [ + 89, + 523, + 482, + 901 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "precise hand representation.", + "bbox": [ + 513, + 92, + 702, + 106 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nc _ {s t r u c} = \\left[ L _ {r} \\left(\\boldsymbol {V} _ {\\boldsymbol {h}}\\right), L _ {r} \\left(\\boldsymbol {J} _ {\\boldsymbol {2 d}}\\right), L _ {n} \\left(\\boldsymbol {T} _ {\\boldsymbol {h}}\\right), L _ {n} \\left(\\boldsymbol {\\theta} _ {\\boldsymbol {h}}\\right) \\right]. \\tag {3}\n$$\n", + "text_format": "latex", + "bbox": [ + 550, + 118, + 906, + 135 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Hand appearance embedding. To improve hand visual detail, we extract features $\\mathcal{F} \\in \\mathbb{R}^{n \\times 1536}$ from the hand Bounding Box regions using DINOv2 [34]. The Hand-Appear processor (HA) uses a linear layer followed by Layer Normalization [1], culminating in the Hand Appearance Embedding $c_{\\text{appear}}$ . This process ensures the hand's visual fidelity and detail are meticulously preserved in our representation.", + "bbox": [ + 511, + 147, + 906, + 267 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nc _ {\\text {a p p e a r}} = \\mathrm {H A} (\\mathcal {F}). \\tag {4}\n$$\n", + "text_format": "latex", + "bbox": [ + 643, + 268, + 906, + 284 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Mask-based regional control. Our method introduces a mask-based regional control to enhance hand detail generation and maintain quality in virtual try-on. Utilizing the hand mask $M_h$ , we modulate Cross-Attention queries, seamlessly integrating Hand Structural Embedding within Hand-Pose Net's Cross-Attention module. Concurrently, Hand Appearance Embedding is adeptly fused into the Try-onNet architecture through mask cross-attention, ensuring precise detail preservation. The mask cross-attention can be represented as:", + "bbox": [ + 511, + 291, + 906, + 441 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\n\\begin{array}{l} Z = \\operatorname {A t t e n t i o n} (Q, K, V, M _ {h}) \\\\ = \\operatorname {S o f t m a x} \\left(\\frac {\\left(Q K ^ {\\top}\\right) \\odot M _ {h}}{\\sqrt {d}}\\right) V, \\tag {5} \\\\ \\end{array}\n$$\n", + "text_format": "latex", + "bbox": [ + 601, + 455, + 905, + 508 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $Q, K$ and $V$ are the query, key, and value matrices of the attention operation for hand appearance embedding cross-attention, $M_h$ represents the hand mask.", + "bbox": [ + 511, + 518, + 905, + 563 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Hand canny shape constraint. During the training process, $\\mathcal{L}_{hand}$ is implemented to refine the clarity and precision of hand edges. Specifically, for a small $R_{t} < T$ , we employ one-step denoising process prior to sampling $x_0$ for the direct prediction of $z_0$ from $z_{t}$ and the model-predicted noise, using Variational Autoencoder for decoder.", + "bbox": [ + 511, + 564, + 905, + 654 + ], + "page_idx": 4 + }, + { + "type": "equation", + "text": "\n$$\nz _ {0} = \\frac {z _ {t} - \\sqrt {1 - \\bar {\\alpha} _ {t}} \\epsilon_ {\\theta}}{\\sqrt {\\bar {\\alpha} _ {t}}}, \\quad t \\leq R _ {t}. \\tag {6}\n$$\n", + "text_format": "latex", + "bbox": [ + 596, + 666, + 906, + 699 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Hand regions are extracted from the generated images using bounding box cropping and are processed with the Canny edge detection operator. These regions are then compared to their corresponding areas in the Ground Truth images using Mean Squared Error (MSE) loss. This loss function is specifically designed to sharpen the model's focus on improving edge details and clarity in the hand regions of the generated images.", + "bbox": [ + 511, + 709, + 905, + 830 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4. Experiment", + "text_level": 1, + "bbox": [ + 513, + 844, + 638, + 861 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Baselines. To comprehensively evaluate our approach, we selected five state-of-the-art diffusion-based methods in the", + "bbox": [ + 511, + 869, + 905, + 898 + ], + "page_idx": 4 + }, + { + "type": "page_number", + "text": "22620", + "bbox": [ + 478, + 944, + 519, + 957 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/87c2702bedbfbc89cfbca50f4f4eb62cfa75b3a211cc8e5d581619b6cc4c3ee6.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 192, + 90, + 803, + 281 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/2331abd65af745254c356932b86a1bea485f1f1ac287398e9bf363e0f013e8f1.jpg", + "image_caption": [ + "(I) Qualitative results on DressCode", + "(II) Qualitative results on VITON-HD", + "Figure 3. Qualitative comparisons of VTON-HandFit with other methods on public datasets: (I) Dresscode and (II) VITON-HD." + ], + "image_footnote": [], + "bbox": [ + 192, + 292, + 803, + 470 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "virtual try-on domain as baselines for comparison: LaDIVTON [32], StableVTON [24], OOTDiffusion [48], IDM-VTON [7], and CatVTON [54]. While these methods have made substantial contributions to the virtual try-on field, they exhibit certain limitations in handling hand occlusion issues, which motivated us to propose VTON-HandFit.", + "bbox": [ + 89, + 515, + 482, + 606 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Dataset. Our experiments utilize two publicly virtual try-on datasets: VITON-HD [6] and DressCode [31]. The VITON-HD dataset consists of 13,679 image pairs, divided into 11,647 for training and 2,032 for testing, featuring front-facing upper-body portraits paired with garments at a resolution of $1024 \\times 768$ . The DressCode dataset, larger and more varied, contains 48,392 training pairs and 5,400 testing pairs, including full-body portraits and a range of garments (upper-body, lower-body, and dresses) at the same resolution. To enhance our evaluation of the model's capability to address hand occlusion challenges, we build Handfit-3K test with 3,620 images sorted from 26,000 images collected in online retail sites.", + "bbox": [ + 89, + 609, + 482, + 806 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Evaluation metrics. Following previous works [24, 32], we adopt both paired and unpaired evaluation settings. In the paired setting, the quality of the generated images is assessed using Structural Similarity (SSIM) [45], Learned Perceptual Image Patch Similarity (LPIPS) [58], Frechet Inception Distance (FID) [18], and Kernel Inception Distance", + "bbox": [ + 89, + 810, + 482, + 901 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "(KID) [3]. For the unpaired setting, we employ FID and KID for comparison. Hand generation quality is assessed using FID-H and KID-H for the hand region. Additionally, we use 2D Mean Per Joint Position Error (MPJPE) [20] to evaluate hand joint position accuracy.", + "bbox": [ + 511, + 515, + 906, + 590 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Implementation details. Our model is developed using the pre-trained StableDiffusion v1.5 [39]. Detailed body dynamics and structural information are captured using DWpose [52] for pose estimation, Detector2 [46] for Dense-sepose segmentation, and HaMeR for detailed hand depth maps and parameters. We use VITON-HD and DressCode datasets for training. Our training strategy is implemented in two phases. In the first phase, we train the Hand-Pose Aggregation Net and Hand-feature Disentanglement Embedding modules while freezing the other components. In the second phase, we finetune the TryonNet and GarmentNet. All models are trained uniformly with a batch size of 32, employing the AdamW [28] optimizer at a learning rate of 5e-5. Training is conducted on 8 NVIDIA V100 GPUs.", + "bbox": [ + 511, + 592, + 908, + 803 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.1. Results on Public Dataset", + "text_level": 1, + "bbox": [ + 511, + 816, + 745, + 830 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Qualitative results. Fig. 3 shows visual comparisons of the VITON-HD and DressCode datasets between VTON-HandFit and other methods. In previous methods, original hand regions are often reused, leading to artifacts such as", + "bbox": [ + 511, + 839, + 906, + 900 + ], + "page_idx": 5 + }, + { + "type": "page_number", + "text": "22621", + "bbox": [ + 478, + 945, + 517, + 955 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/a0ee4e0f278accf5d87660f54d1f89d64502e161d0a37911326d8051d4ea476d.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
MethodsDressCodeVITON-HD
PairedUnpairedPairedUnpaired
SSIM ↑LPIPS ↓FID ↓KID ↓FID ↓KID ↓SSIM ↑LPIPS ↓FID ↓KID ↓FID ↓KID ↓
LaDI-VTON (2023)0.90250.07194.86361.55806.84212.33450.87630.09116.60441.06729.40951.6866
StableVTON (2024)------0.86650.08356.85811.25539.58681.4508
OOTDiffusion (2024)0.89750.07253.94970.71986.70191.86300.85130.09646.51860.89619.67331.2061
IDM-VTON (2024)0.92280.04783.80011.20125.61591.55360.88060.07896.33811.32249.61141.6387
IDM-VTON* (2024)0.90070.06244.91702.01766.68672.57930.83160.10587.74912.33819.77121.9323
CatVTON (2024)0.90110.07053.27550.66965.42201.54900.86940.09706.13940.96399.14341.2666
VTON-HandFit (Ours)0.94040.04442.86660.80095.23141.01890.88500.07305.95640.96998.81780.9951
", + "bbox": [ + 91, + 88, + 903, + 243 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Table 1. Quantitative results on VITON-HD and DressCode datasets. We compare the metrics under both paired (model's clothing is the same as the given cloth image) and unpaired settings (model's clothing differs) with other methods. IDM-VTON* indicates post-processing using HandRefiner.", + "bbox": [ + 89, + 247, + 906, + 289 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/d9fa97f64b899d23b4070c22fbea7fb4d3ec7188994b33d825325e2d3ed392d4.jpg", + "image_caption": [ + "Figure 4. Qualitative comparisons of VTON-HandFit with other methods on our Handfit-3K dataset." + ], + "image_footnote": [], + "bbox": [ + 150, + 297, + 844, + 513 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "blue outlines around the hands when garment colors significantly differ (third row). In contrast, our model effectively eliminates these residual backgrounds and preserves natural hand poses, enhancing visual fidelity and pose accuracy.", + "bbox": [ + 89, + 545, + 483, + 608 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Quantitative results. Tab. 1 presents the quantitative comparisons of VTON-HandFit against other state-of-the-art methods on VITON-HD and DressCode datasets. By enhancing pose accuracy and the fidelity of hand gestures, our approach achieves markedly improved synthesized image quality. Moreover, VTON-HandFit outperforms in FID and KID scores across both datasets in unpaired settings, underscoring the importance of precise hand pose representation for enhancing image quality in practical applications.", + "bbox": [ + 89, + 617, + 483, + 755 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Limitations of handrefiner. Despite the strides made in hand reconstruction through HandRefiner [29], as a post-processing operation, HandRefiner falls short in its inability to reconstruct the hands without complete shape and fails to depict the same ones with the original model accurately. These deficiencies not only undermine the model's practicality but also compromise the authenticity of its output, thereby substantially diminishing its performance metrics and overall applicability, as shown in Tab. 1.", + "bbox": [ + 89, + 763, + 483, + 902 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/dfcf5b3138264a54321f89f0ea28afbeb99987b7953a88108154f40c5f9ae89e.jpg", + "image_caption": [ + "Figure 5. Qualitative comparisons in real-world scenarios." + ], + "image_footnote": [], + "bbox": [ + 529, + 546, + 890, + 681 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/f6b17a074c0e9c0e6efaf9f8797e2ffd923fadad363685b5bcc2457c536c0594.jpg", + "image_caption": [ + "Figure 6. Qualitative comparisons of VTON-HandFit with variants on VITON-HD dataset." + ], + "image_footnote": [], + "bbox": [ + 527, + 705, + 883, + 768 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.2. Results on Handfit-3K Dataset", + "text_level": 1, + "bbox": [ + 511, + 816, + 784, + 830 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Qualitative results. Fig. 4 presents a qualitative comparison between VTON-HandFit and other diffusion-based methods. Our method outperforms in preserving detailed garment attributes and achieving natural hand poses. The", + "bbox": [ + 511, + 839, + 906, + 902 + ], + "page_idx": 6 + }, + { + "type": "page_number", + "text": "22622", + "bbox": [ + 478, + 944, + 519, + 957 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/6d1d9da32f0888bb5f3c5577af8167412fc47605c96aa3c73fa9eab36246fffc.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
MethodsPairedUnpaired
SSIM ↑LPIPS ↓FID ↓KID ↓MPJPE ↓FID-H ↓KID-H ↓FID ↓KID ↓MPJPE ↓FID-H ↓KID-H ↓
LaDI-VTON (2023)0.85970.148012.84304.4112107.631523.019917.455816.25067.0065111.154721.137114.1960
StableVTON (2024)0.87720.102211.36754.136054.758622.747016.69914.04465.196657.333422.809916.3022
IDM-VTON (2024)0.88160.09357.71592.128748.637811.54033.972711.29603.463857.344712.77214.4315
OOTDiffusion (2024)0.85540.11848.71042.136383.965512.19843.665313.19694.483986.167813.80114.4479
CatVTON (2024)0.87610.112310.69664.258160.660813.12644.872315.70816.954463.224515.56035.6134
VTON-HandFit (Ours)0.88270.08195.60270.493225.13767.05741.247410.97863.324826.68929.00032.5362
", + "bbox": [ + 91, + 88, + 903, + 204 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/7ed934ed683f776075d78c6cdfe0ddabd4270fa653ef7d30607c1c6b3db7f1f3.jpg", + "table_caption": [ + "Table 2. Quantitative results on Handfit-3K dataset in the paired and unpaired settings." + ], + "table_footnote": [], + "table_body": "
MethodsSSIM ↑LPIPS ↓FID ↓KID ↓
Baseline0.86190.09236.95491.2651
Baseline+HPA(Densepose)0.86600.08766.89241.1254
Baseline+HPA0.87380.08126.33900.9502
Baseline+HPA+Cstruct0.87840.07876.22251.1186
Baseline+HPA+Cstruct+Cappear0.87870.07926.05151.0084
VTON-HandFit0.88500.07305.95640.9699
", + "bbox": [ + 91, + 239, + 480, + 330 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Table 3. Comparison of different variants on VITON-HD dataset.", + "bbox": [ + 91, + 334, + 480, + 349 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "images generated by VTON-HandFit demonstrate superior consistency in texture, pattern, and fit relative to baseline methods. Furthermore, the hand poses in our images are more consistent with the original.", + "bbox": [ + 89, + 368, + 482, + 428 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Quantitative results. Tab. 2 consolidates the performance of VTON-HandFit against contemporary methods on the Handfit-3K dataset. VTON-HandFit excels in both paired and unpaired settings, showcasing marked advancements in MPJPE, FID-H, and KID-H metrics. This evidences its capability to generate visually compelling images with precise hand pose representation and detailed hand regions. The model's robustness is further affirmed in the unpaired setting, where it consistently leads across all metrics, reinforcing its practical applicability in virtual try-on scenarios with a focus on hand detail fidelity.", + "bbox": [ + 88, + 429, + 482, + 594 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Results of real-world scenarios. To validate the performance of our VTON-HandFit in real-world scenarios, we have collected several cases from natural environments, as illustrated in Fig. 5. It is evident that VTON-HandFit continues to perform robustly in real-world settings, with a noticeable enhancement in hand effects.", + "bbox": [ + 89, + 595, + 482, + 686 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.3. Ablation Study", + "text_level": 1, + "bbox": [ + 89, + 696, + 243, + 712 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Effectiveness of each component. We perform an ablation study to evaluate the contributions of the Hand-Pose Aggregation Net, Hand Structural Embedding $C_{struct}$ , and Hand Appearance Embedding $C_{appear}$ . To facilitate a clear comparison, we introduce several model variants to ascertain the effectiveness of each component. The 'Baseline' model excludes all specialized components. The 'Baseline+HPA (Densepose)' variant incorporates only Densepose for pose control. 'Baseline+HPA' utilizes Densepose, Depth, and DWpose for pose control. 'Baseline+HPA+C_{struct}' excludes the appearance embedding. Lastly, 'Baseline+HPA+C_{struct}+C_{appear}' incorpo", + "bbox": [ + 89, + 719, + 482, + 901 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "rates all components. It's important to note that these variants do not include hand-canny constraint loss. Tab. 3 presents a quantitative comparison of our ablation experiments, highlighting the impact of adding different components on the image generation quality. Fig. 6 offers a qualitative analysis, showing the necessity of pose guidance for accurate arm pose replication in the model. By incorporating stronger hand structure and visual guidance, the generation of hands is significantly improved.", + "bbox": [ + 511, + 242, + 906, + 378 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5. Conclusion", + "text_level": 1, + "bbox": [ + 511, + 393, + 633, + 409 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We propose VTON-HandFit, a novel model designed for addressing hand occlusion in virtual try-on scenario. Specifically, we have developed a Hand-Pose Aggregation Net by explicitly and adaptively encoding global hand and pose priors using the ControlNet-based structure. To fully exploit the hand-related structure and appearance information, we propose the Hand-feature Disentanglement Embedding module, which explicitly separates the hand priors into hand structure-parametric and visual-appearance features. Lastly, we have customized a hand-canny constraint loss to enhance the learning of structural edge knowledge from the hand template of the model image. Our approach has been extensively evaluated qualitatively and quantitatively, demonstrating its superiority over state-of-the-art virtual try-on models, particularly in cases involving hand pose occlusion. These results pave the way for more complex virtual try-on applications toward real-world scenarios.", + "bbox": [ + 511, + 419, + 906, + 678 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Acknowledgments", + "text_level": 1, + "bbox": [ + 511, + 693, + 671, + 709 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "This work was supported by the National Science Fund for Distinguished Young Scholars (No.62025603) and National Natural Science Foundation of China (No. 62072386), supported by Yunnan Provincial Major S&T Special Plan Project (No. 202402AD080001), Henan Key R&D Project (No. 231111212000), Henan Center for Outstanding Overseas Scientists (No. GZS2022011), Open Foundation of Henan Key Lab of General Aviation Technology (No. ZHKF-230212), Key Lab of Oracle Information Processing of MOE (No. OIP2024E002). We also thank Digital Culture Lab of Tencent SSV for their invaluable contribution during our research.", + "bbox": [ + 511, + 719, + 908, + 902 + ], + "page_idx": 7 + }, + { + "type": "page_number", + "text": "22623", + "bbox": [ + 478, + 944, + 517, + 957 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "References", + "text_level": 1, + "bbox": [ + 91, + 89, + 187, + 104 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[1] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint arXiv:1607.06450, 2016. 5", + "[2] Ankan Kumar Bhunia, Salman Khan, Hisham Cholakkal, Rao Muhammad Anwer, Jorma Laaksonen, Mubarak Shah, and Fahad Shahbaz Khan. Person image synthesis via denoising diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5968-5976, 2023. 3", + "[3] Mikołaj Binkowski, Danica J Sutherland, Michael Arbel, and Arthur Gretton. Demystifying mmd gans. arXiv preprint arXiv:1801.01401, 2018. 6", + "[4] Shidong Cao, Wenhao Chai, Shengyu Hao, Yanting Zhang, Hangyue Chen, and Gaoang Wang. Diffashion: Reference-based fashion design with structure-aware transfer by diffusion models. IEEE Transactions on Multimedia, 2023. 3", + "[5] Mengting Chen, Xi Chen, Zhonghua Zhai, Chen Ju, Xuewen Hong, Jinsong Lan, and Shuai Xiao. Wear-any-way: Manipulable virtual try-on via sparse correspondence alignment. arXiv preprint arXiv:2403.12965, 2024. 3", + "[6] Seunghwan Choi, Sunghyun Park, Minsoo Lee, and Jaegul Choo. Viton-hd: High-resolution virtual try-on via misalignment-aware normalization. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14131-14140, 2021. 6", + "[7] Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for virtual try-on. ECCV, 2024. 3, 6", + "[8] Zheng Chong, Xiao Dong, Haoxiang Li, Shiyue Zhang, Wenqing Zhang, Xujie Zhang, Hanqing Zhao, and Xiaodan Liang. Catvton: Concatenation is all you need for virtual try-on with diffusion models. arXiv preprint arXiv:2407.15886, 2024. 1", + "[9] Morelli Davide, Fincato Matteo, Cornia Marcella, Landi Federico, Cesari Fabio, and Cucchiara Rita. Dress code: High-resolution multi-category virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022. 1", + "[10] Haoye Dong, Xiaodan Liang, Xiaohui Shen, Bochao Wang, Hanjiang Lai, Jia Zhu, Zhiting Hu, and Jian Yin. Towards multi-pose guided virtual try-on network. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9026-9035, 2019.", + "[11] Chongjian Ge, Yibing Song, Yuying Ge, Han Yang, Wei Liu, and Ping Luo. Disentangled cycle consistency for highly-realistic virtual try-on. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16928-16937, 2021. 1", + "[12] Yuying Ge, Yibing Song, Ruimao Zhang, Chongjian Ge, Wei Liu, and Ping Luo. Parser-free virtual try-on via distilling appearance flows. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8485-8493, 2021. 3", + "[13] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and" + ], + "bbox": [ + 93, + 114, + 483, + 900 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "Yoshua Bengio. Generative adversarial networks. Communications of the ACM, 63(11):139-144, 2020. 1, 3", + "[14] Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang. Taming the power of diffusion models for high-quality virtual try-on with appearance flow. In Proceedings of the 31st ACM International Conference on Multimedia, pages 7599-7607, 2023. 3", + "[15] Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S Davis. Viton: An image-based virtual try-on network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7543-7552, 2018. 1", + "[16] Xintong Han, Xiaojun Hu, Weilin Huang, and Matthew R Scott. Clothflow: A flow-based model for clothed person generation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 10471-10480, 2019.", + "[17] Sen He, Yi-Zhe Song, and Tao Xiang. Style-based global appearance flow for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3470-3479, 2022. 1", + "[18] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems, 30, 2017. 6", + "[19] Xiaobin Hu, Xu Peng, Donghao Luo, Xiaozhong Ji, Jinlong Peng, Zhengkai Jiang, Jiangning Zhang, Taisong Jin, Chengjie Wang, and Rongrong Ji. Diffumatting: Synthesizing arbitrary objects with matting-level annotation. arXiv preprint arXiv:2403.06168, 2024. 4", + "[20] Catalin Ionescu, Dragos Papava, Vlad Olaru, and Cristian Sminchisescu. Human3. 6m: Large scale datasets and predictive methods for 3d human sensing in natural environments. IEEE transactions on pattern analysis and machine intelligence, 36(7):1325-1339, 2013. 6", + "[21] Thibaut Issenhuth, Jérémie Mary, and Clément Calauzenes. Do not mask what you do not need to mask: a parser-free virtual try-on. In Computer Vision-ECCV 2020: 16th European Conference, Glasgow, UK, August 23-28, 2020, Proceedings, Part XX 16, pages 619-635. Springer, 2020. 3", + "[22] Johanna Karras, Aleksander Holynski, Ting-Chun Wang, and Ira Kemelmacher-Shlizerman. Dreampose: Fashion image-to-video synthesis via stable diffusion. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 22623-22633. IEEE, 2023. 3", + "[23] Bahjat Kawar, Shiran Zada, Oran Lang, Omer Tov, Huiwen Chang, Tali Dekel, Inbar Mosseri, and Michal Irani. Imagic: Text-based real image editing with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6007-6017, 2023. 4", + "[24] Jeongho Kim, Guojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8176-8185, 2024. 3, 6", + "[25] Nupur Kumari, Bingliang Zhang, Richard Zhang, Eli Shechtman, and Jun-Yan Zhu. Multi-concept customization of text-to-image diffusion. In Proceedings of the IEEE/CVF" + ], + "bbox": [ + 517, + 92, + 903, + 900 + ], + "page_idx": 8 + }, + { + "type": "page_number", + "text": "22624", + "bbox": [ + 478, + 945, + 517, + 955 + ], + "page_idx": 8 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "Conference on Computer Vision and Pattern Recognition, pages 1931-1941, 2023. 4", + "[26] Sangyun Lee, Gyojung Gu, Sunghyun Park, Seunghwan Choi, and Jaegul Choo. High-resolution virtual try-on with misalignment and occlusion-handled conditions. In European Conference on Computer Vision, pages 204-219. Springer, 2022. 1, 3", + "[27] Yuhan Li, Hao Zhou, Wenxiang Shang, Ran Lin, Xuanhong Chen, and Bingbing Ni. Anyfit: Controllable virtual try-on for any combination of attire across any scenario. arXiv preprint arXiv:2405.18172, 2024. 3", + "[28] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 6", + "[29] Wenquan Lu, Yufei Xu, Jing Zhang, Chaoyue Wang, and Dacheng Tao. Handrefiner: Refining malformed hands in generated images by diffusion-based conditional inpainting. In ACM Multimedia 2024, 2023. 4, 7", + "[30] Yifang Men, Yiming Mao, Yuning Jiang, Wei-Ying Ma, and Zhouhui Lian. Controllable person image synthesis with attribute-decomposed gan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5084-5093, 2020. 3", + "[31] Davide Morelli, Matteo Fincato, Marcella Cornia, Federico Landi, Fabio Cesari, and Rita Cucchiara. Dress code: High-resolution multi-category virtual try-on. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2231-2235, 2022. 6", + "[32] Davide Morelli, Alberto Baldrati, Giuseppe Cartella, Marcelia Cornia, Marco Bertini, and Rita Cucchiara. Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on. In Proceedings of the 31st ACM International Conference on Multimedia, pages 8580-8589, 2023. 3, 6", + "[33] Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4296–4304, 2024. 4", + "[34] Maxime Oquab, Timothee Darcet, Theo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 5", + "[35] Georgios Pavlakos, Dandan Shan, Ilija Radosavovic, Angjoo Kanazawa, David Fouhey, and Jitendra Malik. Reconstructing hands in 3d with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9826-9836, 2024. 4, 5", + "[36] Anton Pelykh, Ozge Mercanoglu Sincan, and Richard Bowden. Giving a hand to diffusion models: a two-stage approach to improving conditional human image generation. arXiv preprint arXiv:2403.10731, 2024. 4", + "[37] Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952, 2023. 4" + ], + "bbox": [ + 91, + 90, + 482, + 898 + ], + "page_idx": 9 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[38] Sergey Prokudin, Christoph Lassner, and Javier Romero. Efficient learning on point clouds with basis point sets. In Proceedings of the IEEE/CVF international conference on computer vision, pages 4332-4341, 2019. 5", + "[39] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684-10695, 2022. 1, 4, 6", + "[40] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22500-22510, 2023. 4", + "[41] Chitwan Sahara, William Chan, Huiwen Chang, Chris Lee, Jonathan Ho, Tim Salimans, David Fleet, and Mohammad Norouzi. Palette: Image-to-image diffusion models. In ACM SIGGRAPH 2022 conference proceedings, pages 1-10, 2022. 4", + "[42] Dvir Samuel, Rami Ben-Ari, Simon Raviv, Nir Darshan, and Gal Chechik. Generating images of rare concepts using pretrained diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4695-4703, 2024. 4", + "[43] Ke Sun, Jian Cao, Qi Wang, Linrui Tian, Xindi Zhang, Lian Zhuo, Bang Zhang, Liefeng Bo, Wenbo Zhou, Weiming Zhang, et al. Outfitanyone: Ultra-high quality virtual try-on for any clothing and any person. arXiv preprint arXiv:2407.16224, 2024. 3", + "[44] Chengrui Wang, Pengfei Liu, Min Zhou, Ming Zeng, Xubin Li, Tiezheng Ge, et al. Rhand: Refining malformed hands for generated images with decoupled structure and style guidance. arXiv preprint arXiv:2404.13984, 2024. 4", + "[45] Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600-612, 2004. 6", + "[46] Yuxin Wu, Alexander Kirillov, Francisco Massa, Wan-Yen Lo, and Ross Girshick. Detectron2. https://github.com/facebookresearch/detectron2, 2019. 6", + "[47] Zhenyu Xie, Zaiyu Huang, Xin Dong, Fuwei Zhao, Haoye Dong, Xijin Zhang, Feida Zhu, and Xiaodan Liang. Gqvton: Towards general purpose virtual try-on via collaborative local-flow global-parsing learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 23550-23559, 2023. 3", + "[48] Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen. Oot-diffusion: Outfitting fusion based latent diffusion for controllable virtual try-on. arXiv preprint arXiv:2403.01779, 2024.6", + "[49] Binxin Yang, Shuyang Gu, Bo Zhang, Ting Zhang, Xuejin Chen, Xiaoyan Sun, Dong Chen, and Fang Wen. Paint by example: Exemplar-based image editing with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18381-18391, 2023. 3" + ], + "bbox": [ + 516, + 92, + 906, + 898 + ], + "page_idx": 9 + }, + { + "type": "page_number", + "text": "22625", + "bbox": [ + 478, + 945, + 517, + 955 + ], + "page_idx": 9 + }, + { + "type": "list", + "sub_type": "ref_text", + "list_items": [ + "[50] Han Yang, Ruimao Zhang, Xiaobao Guo, Wei Liu, Wangmeng Zuo, and Ping Luo. Towards photo-realistic virtual try-on by adaptively generating-preserving image content. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7850-7859, 2020. 1", + "[51] Zhijing Yang, Junyang Chen, Yukai Shi, Hao Li, Tianshui Chen, and Liang Lin. Occlumix: Towards de-occlusion virtual try-on by semantically-guided mixup. IEEE Transactions on Multimedia, 25:1477–1488, 2023. 3", + "[52] Zhendong Yang, Ailing Zeng, Chun Yuan, and Yu Li. Effective whole-body pose estimation with two-stages distillation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4210-4220, 2023. 6", + "[53] Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv preprint arXiv:2308.06721, 2023.4", + "[54] Jianhao Zeng, Dan Song, Weizhi Nie, Hongshuo Tian, Tongtong Wang, and An-An Liu. Cat-dm: Controllable accelerated virtual try-on with diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8372-8382, 2024. 6", + "[55] Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, Jun Zhu, Lionel M Ni, and Heung-Yeung Shum. Dino: Detr with improved denoising anchor boxes for end-to-end object detection. ICLR, 2023. 2", + "[56] Haozhuo Zhang, Bin Zhu, Yu Cao, and Yanbin Hao. Hand1000: Generating realistic hands from text with only 1,000 images. arXiv preprint arXiv:2408.15461, 2024. 4", + "[57] Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3836-3847, 2023. 4", + "[58] Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 586-595, 2018. 6", + "[59] Xujie Zhang, Ente Lin, Xiu Li, Yuxuan Luo, Michael Kampffmeyer, Xin Dong, and Xiaodan Liang. Mmtryon: Multi-modal multi-reference control for high-quality fashion generation. arXiv preprint arXiv:2405.00448, 2024. 3", + "[60] Shihao Zhao, Dongdong Chen, Yen-Chun Chen, Jianmin Bao, Shaozhe Hao, Lu Yuan, and Kwan-Yee K Wong. Uni-controlnet: All-in-one control to text-to-image diffusion models. Advances in Neural Information Processing Systems, 36, 2024. 4", + "[61] Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Sahara, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two unets. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4606-4615, 2023. 1, 2, 3" + ], + "bbox": [ + 91, + 92, + 480, + 838 + ], + "page_idx": 10 + }, + { + "type": "page_number", + "text": "22626", + "bbox": [ + 478, + 945, + 517, + 955 + ], + "page_idx": 10 + } +] \ No newline at end of file diff --git a/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/00462254-bb4b-4c24-9d0a-9a8f2054757a_model.json b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/00462254-bb4b-4c24-9d0a-9a8f2054757a_model.json new file mode 100644 index 0000000000000000000000000000000000000000..8ffc971bc363bdf8d82caa192f686b845f436d9d --- /dev/null +++ b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/00462254-bb4b-4c24-9d0a-9a8f2054757a_model.json @@ -0,0 +1,1971 @@ +[ + [ + { + "type": "header", + "bbox": [ + 0.107, + 0.003, + 0.182, + 0.043 + ], + "angle": 0, + "content": "CVF" + }, + { + "type": "header", + "bbox": [ + 0.237, + 0.001, + 0.812, + 0.047 + ], + "angle": 0, + "content": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore." + }, + { + "type": "title", + "bbox": [ + 0.125, + 0.13, + 0.877, + 0.178 + ], + "angle": 0, + "content": "VTON-HandFit: Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding" + }, + { + "type": "text", + "bbox": [ + 0.18, + 0.203, + 0.821, + 0.293 + ], + "angle": 0, + "content": "Yujie Liang\\(^{1,2*}\\), Xiaobin Hu\\(^{3*}\\), Boyuan Jiang\\(^{3}\\), Donghao Luo\\(^{3\\dagger}\\), Xu Peng\\(^{3}\\), Kai Wu\\(^{3}\\), Chengming Xu\\(^{3}\\), Wenhui Han\\(^{3}\\), Taisong Jin\\(^{1,2\\dagger}\\), Chengjie Wang\\(^{3}\\), Rongrong Ji\\(^{1,2}\\) \n\\(^{1}\\)Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, China. \n\\(^{2}\\)School of Informatics, Xiamen University, China. \\(^{3}\\)Tencent Youtu Lab." + }, + { + "type": "text", + "bbox": [ + 0.198, + 0.295, + 0.798, + 0.346 + ], + "angle": 0, + "content": "liangyujie@stu.xmu.edu.cn,{xiaobinhu,byronjiang,michaelluo,ppxupeng, chengmingxu,jasoncjwang}@tencent.com,wukaiwork,outlook.com, whhan18@gmail.com,{jintaisong,rrji}@xmu.edu.cn" + }, + { + "type": "title", + "bbox": [ + 0.248, + 0.38, + 0.327, + 0.396 + ], + "angle": 0, + "content": "Abstract" + }, + { + "type": "text", + "bbox": [ + 0.089, + 0.412, + 0.486, + 0.775 + ], + "angle": 0, + "content": "Although diffusion-based image virtual try-on has made considerable progress, emerging approaches still struggle to effectively address the issue of hand occlusion (i.e., clothing regions occluded by the hand part), leading to a notable degradation of the try-on performance. To tackle this issue widely existing in real-world scenarios, we propose VTON-HandFit, leveraging the power of hand priors to reconstruct the appearance and structure for hand occlusion cases. Firstly, we tailor a Hand-Pose Aggregation Net using the ControlNet-based structure explicitly and adaptively encoding the global hand and pose priors. Besides, to fully exploit the hand-related structure and appearance information, we propose Hand-feature Disentanglement Embedding module to disentangle the hand priors into the hand structure-parametric and visual-appearance features, and customize a masked cross attention for further decoupled feature embedding. Lastly, we customize a hand-canny constraint loss to better learn the structure edge knowledge from the hand template of model image. VTON-HandFit outperforms the baselines in qualitative and quantitative evaluations on the public dataset and our self-collected hand-occlusion Handfit-3K dataset particularly for the arbitrary hand pose occlusion cases in real-world scenarios. Our project page is at: https://vton-handfit.github.io." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.8, + 0.224, + 0.815 + ], + "angle": 0, + "content": "1. Introduction" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.825, + 0.484, + 0.871 + ], + "angle": 0, + "content": "Image-based virtual try-on (VTON) [26] is a widely adopted and highly promising image synthesis technology in the e-commerce industry. Its primary goal is to enhance" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.381, + 0.907, + 0.472 + ], + "angle": 0, + "content": "the shopping experience for consumers and minimize advertising expenses for clothing merchants. The VTON task involves generating an image of a human model wearing a specific garment. Over the past few years, countless researchers [9-11, 50] have dedicated significant efforts to achieve more realistic and precise virtual try-on results." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.475, + 0.909, + 0.763 + ], + "angle": 0, + "content": "Most recent studies on virtual try-on predominantly utilize generative adversarial networks (GANs) [13] or latent diffusion models (LDMs) [39] for image synthesis. Traditional GAN-based approaches [15-17] often struggle with accurately generating garment folds, realistic lighting and shadows, and lifelike human body representations. Consequently, more recent research has shifted towards LDM-based methods [61], which effectively improve the authenticity of clothed images. However, these researches focus on generating more realistic and natural try-on images while keeping the detailed features of garments. The hands of try-on models are usually strictly on the sides and garments are not occluded by the hands. Such a strict constraint hinders the virtual try-on paving the way toward real-world applications. For instance, as a highly flexible editing task with a wide range of degrees of freedom, digital humans usually meet the occasions that the garments are occluded by the hands, which severely deteriorates the performance of virtual try-on." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.765, + 0.911, + 0.903 + ], + "angle": 0, + "content": "Motivated by the above challenge, basically, there mainly exist two solutions to tackle this arbitrary hand problem: 1). perfect hand parsing model to accurately segment the tiny area among fingers. 2). involving the hand-reconstruction into the try-on processing. As shown in Fig. 1, CatVTON[8] use the parsing model to keep the hand part, leading to unavoidably the remaining artifacts and backgrounds from the model image. The existing parsing models fail to accurately segment the small" + }, + { + "type": "page_footnote", + "bbox": [ + 0.094, + 0.876, + 0.334, + 0.889 + ], + "angle": 0, + "content": "* equal contribution, † corresponding author" + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.946, + 0.521, + 0.957 + ], + "angle": 0, + "content": "22616" + } + ], + [ + { + "type": "image", + "bbox": [ + 0.111, + 0.089, + 0.885, + 0.405 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.09, + 0.406, + 0.907, + 0.477 + ], + "angle": 0, + "content": "Figure 1. Comparison of different models on virtual try-on with hand occlusion. The leftmost three images are reference images, showing the target clothing, the model, and a close-up of the model's hand. Our model excels in preserving hand details and achieving realistic clothing transfer, as highlighted in the red boxes. CatVTON utilizes a parsing model to retain the hand segment, resulting in the inevitable persistence of residual artifacts and backgrounds from the model image. IDM-VTON* indicates post-processing using HandRefiner, which fails to address missing hands and mistakenly alters correctly depicted ones." + }, + { + "type": "text", + "bbox": [ + 0.093, + 0.49, + 0.485, + 0.897 + ], + "angle": 0, + "content": "and fragmented finger areas, thus we embrace the latter strategy to involve the hand-reconstruction by disentangling the hand into the appearance and structure knowledge processing in the try-on stage. Then, we propose a novel diffusion-based virtual try-on method, termed as VTON-HandFit, to generate realistic and natural images for arbitrary hand position or hand-occlusion cases with the aid of hand 3D structure-parametric and visual-appearance priors. To embed the pose- and hand-related knowledge, we build a Hand-Pose Net to exploit the body and hand semantic and structure information by getting the body (e.g., DWPose, DensPose) and 3D hand HaMeR-based [61] reconstruction information. Secondly, considering the over-small character of hand that is easily ignored after down-sampling operation, we use the DINO [55] to dig the local appearance and texture knowledge of hand as the hand-visual embedding features. For the structure of hand, we adopt these structure-parametric priors (e.g., 3D MANO-related parameters, handtype and hand-box coordinates) as hand-structure embedding features. After getting structure-parametric and visual-appearance features, we aim to enhance the model's ability to respectively learn hand-related patterns. To achieve this, we inject the hand-structure embedding into the Hand-Pose Net with a primary focus on enhancing the latent features related to hand structure, and the hand-visual embedding into the mainstream U-Net specifically aiming to enhance texture-related features, respec" + }, + { + "type": "text", + "bbox": [ + 0.511, + 0.49, + 0.907, + 0.642 + ], + "angle": 0, + "content": "tively. The disentanglement learning of structure and visual appearance enhances the network's ability to efficiently and respectively acquire hand-related knowledge. To prevent the interference of hand-related priors with garment features, we impose a constraint that restricts the hand-related features to operate solely within the hand mask area of the cross-attention mechanism. Finally, we tailor a handcanny constraint loss to more effectively acquire the structural edge knowledge from the hand template of the model image. Our contributions are summarized as follows:" + }, + { + "type": "text", + "bbox": [ + 0.508, + 0.648, + 0.907, + 0.693 + ], + "angle": 0, + "content": "- We propose VTON-HandFit, a novel virtual try-on architecture considering to solve the Hand pose occlusion problem popularly existing in real-world scenarios." + }, + { + "type": "text", + "bbox": [ + 0.508, + 0.698, + 0.907, + 0.773 + ], + "angle": 0, + "content": "- We propose a Hand-Pose Aggregation Net to encode the hand and pose structure priors and also adaptively to adjust the weight of each component. Meanwhile, hand-canny constraint loss is designed to learn the hand structural-edge knowledge of model images." + }, + { + "type": "text", + "bbox": [ + 0.508, + 0.777, + 0.909, + 0.882 + ], + "angle": 0, + "content": "- We propose a novel hand-feature disentanglement embedding module disentangling the hand knowledge into the structure-parametric priors (e.g., MANO parameters) and visual-appearance priors (e.g., color). A mask cross-attention mechanism is proposed to constrain the hand-related features only working on the hand mask area to avoid the garment feature disturbance." + }, + { + "type": "text", + "bbox": [ + 0.508, + 0.886, + 0.907, + 0.9 + ], + "angle": 0, + "content": "- We self-collect Handfit-3K test dataset from online retail" + }, + { + "type": "list", + "bbox": [ + 0.508, + 0.648, + 0.909, + 0.9 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.521, + 0.958 + ], + "angle": 0, + "content": "22617" + } + ], + [ + { + "type": "image", + "bbox": [ + 0.144, + 0.09, + 0.852, + 0.449 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.089, + 0.456, + 0.908, + 0.553 + ], + "angle": 0, + "content": "Figure 2. An overview of our VTON-HandFit Network. The network consists of two main components: Hand-feature Disentanglement Embedding and Hand-Pose Aggregation Net. The Hand-feature Disentanglement Embedding module uses the HaMer model to extract hand priors, including hand type \\( T_{h} \\), 3D vertices \\( V_{h} \\), spatial joint locations \\( J_{2d} \\), and joint rotation matrices \\( \\theta_{h} \\). These features are processed by the Hand-Struct processor to derive structural features \\( c_{struc} \\). Simultaneously, hand images cropped using bounding boxes are processed by DINOv2 and the Hand-Appear processor to obtain visual features \\( c_{appear} \\). The structural and visual features are integrated using mask cross attention. The Hand-Pose Aggregation Net module controls body and hand poses by aggregating DWpose, Densepose, and hand depth maps." + }, + { + "type": "text", + "bbox": [ + 0.098, + 0.567, + 0.484, + 0.645 + ], + "angle": 0, + "content": "sites to evaluate hand occlusion performance. Extensive qualitative and quantitative evaluations demonstrate our superiority over state-of-the-art virtual try-on models especially for hand pose occlusion cases, paving the road towards more complicated virtual try-on in real-world." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.662, + 0.241, + 0.679 + ], + "angle": 0, + "content": "2. Related Works" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.689, + 0.484, + 0.901 + ], + "angle": 0, + "content": "Image-based virtual try-on. In the context of image-based virtual try-on, the objective is to generate a realistic representation of a target person wearing a specific garment, given a pair of corresponding images. A series of studies [26, 30, 47, 51] have been conducted using Generative Adversarial Networks (GANs) [13]. These studies typically involve the initial deformation of the garment to match the body shape of the person, followed by the utilization of a generator to superimpose the deformed garment onto the person image. Despite efforts made to minimize the discrepancy between the warped garment and the person [12, 21, 26], these approaches often lack the ability to generalize well to diverse person images, particularly those with complex backgrounds or intricate poses. There exists" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.567, + 0.907, + 0.628 + ], + "angle": 0, + "content": "GAN-based try-on method [26] considers occluded problems and designs the warping operation to avoid the garment warping misalignment, which is totally different from diffusion-based algorithms as an inpainting problem." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.63, + 0.909, + 0.903 + ], + "angle": 0, + "content": "The remarkable generative capabilities of diffusion models have sparked interest in incorporating them into fashion synthesis, particularly in tasks like visual try-on [2, 4, 5, 22, 27, 43, 59]. TryOnDiffusion [61] employs ParallelUNets to preserve garment details and warp the garment for the try-on process. Subsequent studies have treated virtual try-on as an exemplar-based image inpainting problem [49]. These studies involved fine-tuning the inpainting diffusion models using virtual try-on datasets [24, 32] to generate virtual try-on images of superior quality. For instance, such as LADI-VTON [32] and DCI-VTON [14] have emerged, treating clothing as pseudo-words or employing warping networks to integrate garments into pre-trained diffusion models. IDM-VTON [7] incorporates sophisticated attention modules to encode the high-level semantics of the garment and extract low-level features, thereby preserving fine-grained details. While these diffusion methods address issues related to the nature and realism of synthesized images," + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.946, + 0.521, + 0.957 + ], + "angle": 0, + "content": "22618" + } + ], + [ + { + "type": "text", + "bbox": [ + 0.09, + 0.092, + 0.482, + 0.136 + ], + "angle": 0, + "content": "they struggle to meet hard occasions of the hand-based occlusion which very widely exists in the real-world virtual try-on scenarios." + }, + { + "type": "text", + "bbox": [ + 0.093, + 0.138, + 0.483, + 0.453 + ], + "angle": 0, + "content": "Conditional control of diffusion models. In recent years, latent diffusion models [39] have demonstrated significant success in tasks such as text-to-image generation [19, 25, 37, 40] and image-to-image generation [23, 41]. To achieve more controllable generated results, several studies have introduced conditional control to diffusion models [53, 60]. ControlNet [57] adds spatial conditioning controls (e.g.,anny edges, Hough lines, user scribbles, human key points) to pretrained diffusion models via zero convolutions manner. T2I-Adapter [33] aligns the internal knowledge within T2I models with external control signals, allowing it to be trained based on different conditions. This approach enables the generation of results with rich control and editing effects, particularly in terms of color and structure. In our paper, we focus on the hard occasions of hand-based occlusion in image-based VTON task, and tailor a Hand-Pose Aggregation Net to control the body pose and hand structure. Furthermore, the disengagement knowledge of hand structure and appearance embedding is studied and then these embeddings are respectively injected into the attention mechanism, acting as a mask-based regional control." + }, + { + "type": "text", + "bbox": [ + 0.093, + 0.455, + 0.483, + 0.786 + ], + "angle": 0, + "content": "Reliable human hand generation. Previous studies [42] have acknowledged the problem of misshaped hands that are generated by existing text-to-image diffusion models. Controlling the hand is considerably more challenging compared to controlling the body pose, primarily due to the relative size of the main subject. Besides, diffusion models exhibit significantly improved capability in generating plausible whole-body poses. This difficulty arises from the intricate nature of learning the physical structure and pose of hands from training images, which encompasses extensive deformations and occlusions. Handrefiner [29] and RHanDS [44] are post-processing solutions adopting a hand mesh reconstruction model to rectify malformed hands. Hand-to-Diffusion [36] uses a two-stage approach, first generating hand effects and then producing the entire image. Hand1000 [56] employs a three-stage training process that integrates hand gesture recognition and optimized text embeddings. VTON-HandFit involves the hand corrections into the generation process not as a post-processing and disentangles the hand priors into the structure and appearance feature embedding for virtual try-on arbitrary hand pose in the real-world." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.8, + 0.226, + 0.817 + ], + "angle": 0, + "content": "3. Methodology" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.825, + 0.482, + 0.901 + ], + "angle": 0, + "content": "Data collection of Handfit-3K. Considering that the hand of the current main virtual try-on dataset is separate from the body, there rarely exists a great amount of virtual try-on data for arbitrary hand pose and hard hand-occlusion cases. To meet this urgent need in real-world, we aim to" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.092, + 0.905, + 0.273 + ], + "angle": 0, + "content": "construct a dataset of complex hand poses that occlude the garment. Specifically, we collect an additional dataset of 26,000 images sourced from online retail sites. These images underwent processing using a parsing model and a hand bounding box via HaMeR model [35] to identify instances of occlusion. We devote significant manual effort to inspecting the quality of the data, removing any erroneous or flawed entries, thereby ensuring the high quality of the virtual try-on dataset. As a result, we obtained a specialized hand occlusion test set comprising 3,620 images, termed as Handfit-3K. This dataset facilitates a targeted evaluation of our method's ability to effectively handle occlusions." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.282, + 0.644, + 0.298 + ], + "angle": 0, + "content": "3.1. Preliminary" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.304, + 0.905, + 0.395 + ], + "angle": 0, + "content": "Latent Diffusion Models. Our VTON-HandFit is an extension of stable diffusion model, which is a probabilistic model specifically designed to estimate a data distribution \\( \\mathfrak{p}(\\mathbf{x}) \\) by iteratively reducing noise in a normally distributed variable. Given an input image \\( x_0 \\), the noise estimation process is:" + }, + { + "type": "equation", + "bbox": [ + 0.522, + 0.404, + 0.905, + 0.423 + ], + "angle": 0, + "content": "\\[\n\\mathcal {L} _ {\\text {n o i s e}} = \\mathbb {E} _ {z \\sim \\mathcal {E} (x), C, \\epsilon \\sim \\mathcal {N} (0, 1), t} \\left[ | | \\epsilon - \\epsilon_ {\\theta} (z _ {t}, t, C) | | _ {2} ^ {2} \\right], \\tag {1}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.431, + 0.905, + 0.507 + ], + "angle": 0, + "content": "where a Variational AutoEncoder \\(\\mathcal{E}\\) compress input image \\(x\\) into low-dimension latent space \\(z\\). A conditional U-Net denoiser \\(\\epsilon_{\\theta}\\) aims to estimate noise \\(\\theta\\) in latent space with the aid of timestep \\(t\\), \\(t\\)-th noisy latent representation \\(z_{t}\\), and other text-prompt conditions \\(C\\) extracted by text-encoder." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.515, + 0.673, + 0.53 + ], + "angle": 0, + "content": "3.2.VTON-HandFit" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.539, + 0.905, + 0.901 + ], + "angle": 0, + "content": "Overview. An overview of our VTON-HandFit is illustrated in Fig. 2. Given a target person image \\( x \\in \\mathbb{R}^{H \\times W \\times 3} \\) and an input garment image \\( g \\in \\mathbb{R}^{H \\times W \\times 3} \\), VTON-HandFit aims to generate a realistic outfitted image \\( x_{g} \\in \\mathbb{R}^{H \\times W \\times 3} \\). The garment-agnostic person representation \\( x_{m} \\) is masked to erase the garment information by the combination of HumanParsing and HumanPose. Thus, this virtual try-on task is transformed into an example-based inpainting problem to recover the missing mask part. For the input of TryonNet, we simply concatenate the noise image \\( (z_{t}) \\) and latent garment-agnostic features \\( \\mathcal{E}(x_{m}) \\) to avoid the other pose-relevant information disturbances. In Hand-Pose Aggregation Net, we also aggregate the pose-related knowledge including the DWPose, DensePose and 3D reconstruction depth map derived by the HaMeR mesh model. These pose-related priors are detached with TryonNet to explicitly be learned in Hand-Pose Net. DensePose provides more bodysemantic information, DWPose contains abundant skeleton structure, and 3D-hand-derived depth maps demonstrate the pixel-level global-aligned hand position and structure knowledge. These modules complement each other by mutually providing information, and working together to constrain the pose and hand generation. In hand-feature disentanglement embedding, we use the HaMeR model for 3D" + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.945, + 0.519, + 0.957 + ], + "angle": 0, + "content": "22619" + } + ], + [ + { + "type": "text", + "bbox": [ + 0.09, + 0.092, + 0.482, + 0.227 + ], + "angle": 0, + "content": "hand reconstruction and get the structure-parametric priors. These features are processed by the Hand-Struct processor to derive structural features \\( c_{struc} \\). Simultaneously, cropped hand images are processed by DINOv2 and the Hand-Appear processor to obtain visual features \\( c_{appear} \\). Besides, GarmentNet aims to extract the garment features into the TryonNet mainly by: 1). encoding the garment image into the latent space \\( \\mathcal{E}_g \\), 2). using the CLIP to get the image token features for injection." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.228, + 0.482, + 0.364 + ], + "angle": 0, + "content": "Hand-Pose Aggregation Net. To aggregate the hand and pose-relevant priors learning, we tailor a Hand-Pose Net to constrain the semantic and skeleton pose, and global pixel-level hand knowledge. Specifically, dwpose, densepose, and render blocks are designed based on the same zero-multilayer structure with different weights to respectively extract the corresponding features. Zero-multilayer structure consists of seven stacking of conv and SiLU activate function with the zero convolution layer at the last layer." + }, + { + "type": "equation", + "bbox": [ + 0.167, + 0.376, + 0.483, + 0.392 + ], + "angle": 0, + "content": "\\[\n\\mathcal {F} _ {h p} = \\mathcal {F} _ {d w} + \\mathcal {F} _ {d p} + \\mathcal {F} _ {r h} \\times w _ {\\text {h a n d}}, \\tag {2}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.402, + 0.483, + 0.522 + ], + "angle": 0, + "content": "where \\(\\mathcal{F}_{dw}\\), \\(\\mathcal{F}_{dp}\\), \\(\\mathcal{F}_{rh}\\) are the DWPose feature, Densepose feature, and hand feature after the Zero-multilayer structure, and then the summation is implemented on the pixel-level with the adjustable weight \\(w_{\\mathrm{hand}}\\) of hand feature \\(\\mathcal{F}_{rh}\\). To mitigate the impact of detrimental noise on the hidden states of the neural network layers in the trainable Hand-Pose Net during the initial stages of training, we also adopt the zero convolution learning paradigm of ControlNet." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.524, + 0.483, + 0.902 + ], + "angle": 0, + "content": "Hand structural embedding. The HaMeR [35] model is utilized to reconstruct various MANO parameters and other hand-related information from 2D images, including hand mask \\( M_h \\in \\mathbb{R}^{768 \\times 1024} \\), hand types \\( T_h \\in \\mathbb{R}^{n \\times 1} \\), hand 3D vertices \\( V_h \\in \\mathbb{R}^{n \\times 778 \\times 3} \\), spatial joint locations \\( J_{2d} \\in \\mathbb{R}^{n \\times 21 \\times 2} \\), and joint rotation matrices \\( \\theta \\in \\mathbb{R}^{n \\times 16 \\times 3 \\times 3} \\) where \\( n \\) represents the number of hands. To accommodate variable numbers of hands as input, we pad the hand parameters to a fixed length \\( n = 2 \\) by adding padding at the end. For hand type \\( T_h \\), we use 0 to denote left hands, 1 for right hands, and -1 as a filler for images without hands. Similar padding is applied to other parameters to ensure alignment and correspondence across all dimensions. We apply Basis Point Set (BPS) [38] for dimensionality reduction of hand vertices and encode joint rotations as 6D vectors. And then, we import theses information as well as joint rotations and hand types into the hand-structure processor. The structure of hand-structure processor consists of two parallel blocks: three stacked layers of linear and ReLU activation \\( (L_r) \\) and two stacked layers of linear and layer normalization \\( (L_n) \\). These streamlined hand vertices and joint locations are then fed into \\( L_r \\). Concurrently, joint rotations and hand types are transformed via \\( L_n \\). The outputs from these processes are aggregated to form the Hand Structural Embedding \\( c_{struc} \\), encapsulating essential geometric and 3D information for" + }, + { + "type": "text", + "bbox": [ + 0.514, + 0.093, + 0.704, + 0.107 + ], + "angle": 0, + "content": "precise hand representation." + }, + { + "type": "equation", + "bbox": [ + 0.552, + 0.119, + 0.907, + 0.136 + ], + "angle": 0, + "content": "\\[\nc _ {s t r u c} = \\left[ L _ {r} \\left(\\boldsymbol {V} _ {\\boldsymbol {h}}\\right), L _ {r} \\left(\\boldsymbol {J} _ {\\boldsymbol {2 d}}\\right), L _ {n} \\left(\\boldsymbol {T} _ {\\boldsymbol {h}}\\right), L _ {n} \\left(\\boldsymbol {\\theta} _ {\\boldsymbol {h}}\\right) \\right]. \\tag {3}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.148, + 0.907, + 0.268 + ], + "angle": 0, + "content": "Hand appearance embedding. To improve hand visual detail, we extract features \\(\\mathcal{F} \\in \\mathbb{R}^{n \\times 1536}\\) from the hand Bounding Box regions using DINOv2 [34]. The Hand-Appear processor (HA) uses a linear layer followed by Layer Normalization [1], culminating in the Hand Appearance Embedding \\(c_{\\text{appear}}\\). This process ensures the hand's visual fidelity and detail are meticulously preserved in our representation." + }, + { + "type": "equation", + "bbox": [ + 0.644, + 0.269, + 0.907, + 0.285 + ], + "angle": 0, + "content": "\\[\nc _ {\\text {a p p e a r}} = \\mathrm {H A} (\\mathcal {F}). \\tag {4}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.292, + 0.907, + 0.443 + ], + "angle": 0, + "content": "Mask-based regional control. Our method introduces a mask-based regional control to enhance hand detail generation and maintain quality in virtual try-on. Utilizing the hand mask \\( M_h \\), we modulate Cross-Attention queries, seamlessly integrating Hand Structural Embedding within Hand-Pose Net's Cross-Attention module. Concurrently, Hand Appearance Embedding is adeptly fused into the Try-onNet architecture through mask cross-attention, ensuring precise detail preservation. The mask cross-attention can be represented as:" + }, + { + "type": "equation", + "bbox": [ + 0.602, + 0.456, + 0.906, + 0.509 + ], + "angle": 0, + "content": "\\[\n\\begin{array}{l} Z = \\operatorname {A t t e n t i o n} (Q, K, V, M _ {h}) \\\\ = \\operatorname {S o f t m a x} \\left(\\frac {\\left(Q K ^ {\\top}\\right) \\odot M _ {h}}{\\sqrt {d}}\\right) V, \\tag {5} \\\\ \\end{array}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.519, + 0.906, + 0.564 + ], + "angle": 0, + "content": "where \\( Q, K \\) and \\( V \\) are the query, key, and value matrices of the attention operation for hand appearance embedding cross-attention, \\( M_h \\) represents the hand mask." + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.565, + 0.906, + 0.655 + ], + "angle": 0, + "content": "Hand canny shape constraint. During the training process, \\(\\mathcal{L}_{hand}\\) is implemented to refine the clarity and precision of hand edges. Specifically, for a small \\(R_{t} < T\\), we employ one-step denoising process prior to sampling \\(x_0\\) for the direct prediction of \\(z_0\\) from \\(z_{t}\\) and the model-predicted noise, using Variational Autoencoder for decoder." + }, + { + "type": "equation", + "bbox": [ + 0.598, + 0.667, + 0.907, + 0.7 + ], + "angle": 0, + "content": "\\[\nz _ {0} = \\frac {z _ {t} - \\sqrt {1 - \\bar {\\alpha} _ {t}} \\epsilon_ {\\theta}}{\\sqrt {\\bar {\\alpha} _ {t}}}, \\quad t \\leq R _ {t}. \\tag {6}\n\\]" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.71, + 0.906, + 0.831 + ], + "angle": 0, + "content": "Hand regions are extracted from the generated images using bounding box cropping and are processed with the Canny edge detection operator. These regions are then compared to their corresponding areas in the Ground Truth images using Mean Squared Error (MSE) loss. This loss function is specifically designed to sharpen the model's focus on improving edge details and clarity in the hand regions of the generated images." + }, + { + "type": "title", + "bbox": [ + 0.514, + 0.845, + 0.64, + 0.862 + ], + "angle": 0, + "content": "4. Experiment" + }, + { + "type": "text", + "bbox": [ + 0.513, + 0.871, + 0.906, + 0.9 + ], + "angle": 0, + "content": "Baselines. To comprehensively evaluate our approach, we selected five state-of-the-art diffusion-based methods in the" + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.945, + 0.52, + 0.958 + ], + "angle": 0, + "content": "22620" + } + ], + [ + { + "type": "image", + "bbox": [ + 0.193, + 0.092, + 0.805, + 0.282 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.415, + 0.283, + 0.593, + 0.294 + ], + "angle": 0, + "content": "(I) Qualitative results on DressCode" + }, + { + "type": "image", + "bbox": [ + 0.194, + 0.293, + 0.804, + 0.472 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.408, + 0.473, + 0.593, + 0.485 + ], + "angle": 0, + "content": "(II) Qualitative results on VITON-HD" + }, + { + "type": "image_caption", + "bbox": [ + 0.119, + 0.489, + 0.877, + 0.505 + ], + "angle": 0, + "content": "Figure 3. Qualitative comparisons of VTON-HandFit with other methods on public datasets: (I) Dresscode and (II) VITON-HD." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.516, + 0.483, + 0.607 + ], + "angle": 0, + "content": "virtual try-on domain as baselines for comparison: LaDIVTON [32], StableVTON [24], OOTDiffusion [48], IDM-VTON [7], and CatVTON [54]. While these methods have made substantial contributions to the virtual try-on field, they exhibit certain limitations in handling hand occlusion issues, which motivated us to propose VTON-HandFit." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.611, + 0.483, + 0.807 + ], + "angle": 0, + "content": "Dataset. Our experiments utilize two publicly virtual try-on datasets: VITON-HD [6] and DressCode [31]. The VITON-HD dataset consists of 13,679 image pairs, divided into 11,647 for training and 2,032 for testing, featuring front-facing upper-body portraits paired with garments at a resolution of \\(1024 \\times 768\\). The DressCode dataset, larger and more varied, contains 48,392 training pairs and 5,400 testing pairs, including full-body portraits and a range of garments (upper-body, lower-body, and dresses) at the same resolution. To enhance our evaluation of the model's capability to address hand occlusion challenges, we build Handfit-3K test with 3,620 images sorted from 26,000 images collected in online retail sites." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.811, + 0.483, + 0.902 + ], + "angle": 0, + "content": "Evaluation metrics. Following previous works [24, 32], we adopt both paired and unpaired evaluation settings. In the paired setting, the quality of the generated images is assessed using Structural Similarity (SSIM) [45], Learned Perceptual Image Patch Similarity (LPIPS) [58], Frechet Inception Distance (FID) [18], and Kernel Inception Distance" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.516, + 0.907, + 0.592 + ], + "angle": 0, + "content": "(KID) [3]. For the unpaired setting, we employ FID and KID for comparison. Hand generation quality is assessed using FID-H and KID-H for the hand region. Additionally, we use 2D Mean Per Joint Position Error (MPJPE) [20] to evaluate hand joint position accuracy." + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.593, + 0.909, + 0.804 + ], + "angle": 0, + "content": "Implementation details. Our model is developed using the pre-trained StableDiffusion v1.5 [39]. Detailed body dynamics and structural information are captured using DWpose [52] for pose estimation, Detector2 [46] for Dense-sepose segmentation, and HaMeR for detailed hand depth maps and parameters. We use VITON-HD and DressCode datasets for training. Our training strategy is implemented in two phases. In the first phase, we train the Hand-Pose Aggregation Net and Hand-feature Disentanglement Embedding modules while freezing the other components. In the second phase, we finetune the TryonNet and GarmentNet. All models are trained uniformly with a batch size of 32, employing the AdamW [28] optimizer at a learning rate of 5e-5. Training is conducted on 8 NVIDIA V100 GPUs." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.817, + 0.746, + 0.831 + ], + "angle": 0, + "content": "4.1. Results on Public Dataset" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.84, + 0.908, + 0.901 + ], + "angle": 0, + "content": "Qualitative results. Fig. 3 shows visual comparisons of the VITON-HD and DressCode datasets between VTON-HandFit and other methods. In previous methods, original hand regions are often reused, leading to artifacts such as" + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.946, + 0.518, + 0.957 + ], + "angle": 0, + "content": "22621" + } + ], + [ + { + "type": "table", + "bbox": [ + 0.093, + 0.089, + 0.904, + 0.244 + ], + "angle": 0, + "content": "
MethodsDressCodeVITON-HD
PairedUnpairedPairedUnpaired
SSIM ↑LPIPS ↓FID ↓KID ↓FID ↓KID ↓SSIM ↑LPIPS ↓FID ↓KID ↓FID ↓KID ↓
LaDI-VTON (2023)0.90250.07194.86361.55806.84212.33450.87630.09116.60441.06729.40951.6866
StableVTON (2024)------0.86650.08356.85811.25539.58681.4508
OOTDiffusion (2024)0.89750.07253.94970.71986.70191.86300.85130.09646.51860.89619.67331.2061
IDM-VTON (2024)0.92280.04783.80011.20125.61591.55360.88060.07896.33811.32249.61141.6387
IDM-VTON* (2024)0.90070.06244.91702.01766.68672.57930.83160.10587.74912.33819.77121.9323
CatVTON (2024)0.90110.07053.27550.66965.42201.54900.86940.09706.13940.96399.14341.2666
VTON-HandFit (Ours)0.94040.04442.86660.80095.23141.01890.88500.07305.95640.96998.81780.9951
" + }, + { + "type": "table_caption", + "bbox": [ + 0.09, + 0.248, + 0.907, + 0.29 + ], + "angle": 0, + "content": "Table 1. Quantitative results on VITON-HD and DressCode datasets. We compare the metrics under both paired (model's clothing is the same as the given cloth image) and unpaired settings (model's clothing differs) with other methods. IDM-VTON* indicates post-processing using HandRefiner." + }, + { + "type": "image", + "bbox": [ + 0.151, + 0.298, + 0.845, + 0.515 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.2, + 0.52, + 0.796, + 0.535 + ], + "angle": 0, + "content": "Figure 4. Qualitative comparisons of VTON-HandFit with other methods on our Handfit-3K dataset." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.546, + 0.484, + 0.609 + ], + "angle": 0, + "content": "blue outlines around the hands when garment colors significantly differ (third row). In contrast, our model effectively eliminates these residual backgrounds and preserves natural hand poses, enhancing visual fidelity and pose accuracy." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.618, + 0.484, + 0.756 + ], + "angle": 0, + "content": "Quantitative results. Tab. 1 presents the quantitative comparisons of VTON-HandFit against other state-of-the-art methods on VITON-HD and DressCode datasets. By enhancing pose accuracy and the fidelity of hand gestures, our approach achieves markedly improved synthesized image quality. Moreover, VTON-HandFit outperforms in FID and KID scores across both datasets in unpaired settings, underscoring the importance of precise hand pose representation for enhancing image quality in practical applications." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.765, + 0.484, + 0.903 + ], + "angle": 0, + "content": "Limitations of handrefiner. Despite the strides made in hand reconstruction through HandRefiner [29], as a post-processing operation, HandRefiner falls short in its inability to reconstruct the hands without complete shape and fails to depict the same ones with the original model accurately. These deficiencies not only undermine the model's practicality but also compromise the authenticity of its output, thereby substantially diminishing its performance metrics and overall applicability, as shown in Tab. 1." + }, + { + "type": "image", + "bbox": [ + 0.53, + 0.547, + 0.892, + 0.683 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.536, + 0.687, + 0.882, + 0.701 + ], + "angle": 0, + "content": "Figure 5. Qualitative comparisons in real-world scenarios." + }, + { + "type": "image", + "bbox": [ + 0.528, + 0.707, + 0.885, + 0.77 + ], + "angle": 0, + "content": null + }, + { + "type": "image_caption", + "bbox": [ + 0.513, + 0.776, + 0.907, + 0.804 + ], + "angle": 0, + "content": "Figure 6. Qualitative comparisons of VTON-HandFit with variants on VITON-HD dataset." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.817, + 0.785, + 0.832 + ], + "angle": 0, + "content": "4.2. Results on Handfit-3K Dataset" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.84, + 0.907, + 0.903 + ], + "angle": 0, + "content": "Qualitative results. Fig. 4 presents a qualitative comparison between VTON-HandFit and other diffusion-based methods. Our method outperforms in preserving detailed garment attributes and achieving natural hand poses. The" + }, + { + "type": "page_number", + "bbox": [ + 0.479, + 0.945, + 0.521, + 0.958 + ], + "angle": 0, + "content": "22622" + } + ], + [ + { + "type": "table", + "bbox": [ + 0.093, + 0.089, + 0.905, + 0.205 + ], + "angle": 0, + "content": "
MethodsPairedUnpaired
SSIM ↑LPIPS ↓FID ↓KID ↓MPJPE ↓FID-H ↓KID-H ↓FID ↓KID ↓MPJPE ↓FID-H ↓KID-H ↓
LaDI-VTON (2023)0.85970.148012.84304.4112107.631523.019917.455816.25067.0065111.154721.137114.1960
StableVTON (2024)0.87720.102211.36754.136054.758622.747016.69914.04465.196657.333422.809916.3022
IDM-VTON (2024)0.88160.09357.71592.128748.637811.54033.972711.29603.463857.344712.77214.4315
OOTDiffusion (2024)0.85540.11848.71042.136383.965512.19843.665313.19694.483986.167813.80114.4479
CatVTON (2024)0.87610.112310.69664.258160.660813.12644.872315.70816.954463.224515.56035.6134
VTON-HandFit (Ours)0.88270.08195.60270.493225.13767.05741.247410.97863.324826.68929.00032.5362
" + }, + { + "type": "table_caption", + "bbox": [ + 0.241, + 0.209, + 0.755, + 0.224 + ], + "angle": 0, + "content": "Table 2. Quantitative results on Handfit-3K dataset in the paired and unpaired settings." + }, + { + "type": "table", + "bbox": [ + 0.093, + 0.24, + 0.481, + 0.332 + ], + "angle": 0, + "content": "
MethodsSSIM ↑LPIPS ↓FID ↓KID ↓
Baseline0.86190.09236.95491.2651
Baseline+HPA(Densepose)0.86600.08766.89241.1254
Baseline+HPA0.87380.08126.33900.9502
Baseline+HPA+Cstruct0.87840.07876.22251.1186
Baseline+HPA+Cstruct+Cappear0.87870.07926.05151.0084
VTON-HandFit0.88500.07305.95640.9699
" + }, + { + "type": "table_caption", + "bbox": [ + 0.092, + 0.335, + 0.482, + 0.35 + ], + "angle": 0, + "content": "Table 3. Comparison of different variants on VITON-HD dataset." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.369, + 0.483, + 0.429 + ], + "angle": 0, + "content": "images generated by VTON-HandFit demonstrate superior consistency in texture, pattern, and fit relative to baseline methods. Furthermore, the hand poses in our images are more consistent with the original." + }, + { + "type": "text", + "bbox": [ + 0.089, + 0.43, + 0.483, + 0.595 + ], + "angle": 0, + "content": "Quantitative results. Tab. 2 consolidates the performance of VTON-HandFit against contemporary methods on the Handfit-3K dataset. VTON-HandFit excels in both paired and unpaired settings, showcasing marked advancements in MPJPE, FID-H, and KID-H metrics. This evidences its capability to generate visually compelling images with precise hand pose representation and detailed hand regions. The model's robustness is further affirmed in the unpaired setting, where it consistently leads across all metrics, reinforcing its practical applicability in virtual try-on scenarios with a focus on hand detail fidelity." + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.596, + 0.483, + 0.687 + ], + "angle": 0, + "content": "Results of real-world scenarios. To validate the performance of our VTON-HandFit in real-world scenarios, we have collected several cases from natural environments, as illustrated in Fig. 5. It is evident that VTON-HandFit continues to perform robustly in real-world settings, with a noticeable enhancement in hand effects." + }, + { + "type": "title", + "bbox": [ + 0.091, + 0.697, + 0.245, + 0.713 + ], + "angle": 0, + "content": "4.3. Ablation Study" + }, + { + "type": "text", + "bbox": [ + 0.09, + 0.72, + 0.483, + 0.902 + ], + "angle": 0, + "content": "Effectiveness of each component. We perform an ablation study to evaluate the contributions of the Hand-Pose Aggregation Net, Hand Structural Embedding \\( C_{struct} \\), and Hand Appearance Embedding \\( C_{appear} \\). To facilitate a clear comparison, we introduce several model variants to ascertain the effectiveness of each component. The 'Baseline' model excludes all specialized components. The 'Baseline+HPA (Densepose)' variant incorporates only Densepose for pose control. 'Baseline+HPA' utilizes Densepose, Depth, and DWpose for pose control. 'Baseline+HPA+C_{struct}' excludes the appearance embedding. Lastly, 'Baseline+HPA+C_{struct}+C_{appear}' incorpo" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.243, + 0.907, + 0.38 + ], + "angle": 0, + "content": "rates all components. It's important to note that these variants do not include hand-canny constraint loss. Tab. 3 presents a quantitative comparison of our ablation experiments, highlighting the impact of adding different components on the image generation quality. Fig. 6 offers a qualitative analysis, showing the necessity of pose guidance for accurate arm pose replication in the model. By incorporating stronger hand structure and visual guidance, the generation of hands is significantly improved." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.395, + 0.634, + 0.41 + ], + "angle": 0, + "content": "5. Conclusion" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.42, + 0.907, + 0.679 + ], + "angle": 0, + "content": "We propose VTON-HandFit, a novel model designed for addressing hand occlusion in virtual try-on scenario. Specifically, we have developed a Hand-Pose Aggregation Net by explicitly and adaptively encoding global hand and pose priors using the ControlNet-based structure. To fully exploit the hand-related structure and appearance information, we propose the Hand-feature Disentanglement Embedding module, which explicitly separates the hand priors into hand structure-parametric and visual-appearance features. Lastly, we have customized a hand-canny constraint loss to enhance the learning of structural edge knowledge from the hand template of the model image. Our approach has been extensively evaluated qualitatively and quantitatively, demonstrating its superiority over state-of-the-art virtual try-on models, particularly in cases involving hand pose occlusion. These results pave the way for more complex virtual try-on applications toward real-world scenarios." + }, + { + "type": "title", + "bbox": [ + 0.513, + 0.694, + 0.673, + 0.71 + ], + "angle": 0, + "content": "Acknowledgments" + }, + { + "type": "text", + "bbox": [ + 0.512, + 0.72, + 0.909, + 0.903 + ], + "angle": 0, + "content": "This work was supported by the National Science Fund for Distinguished Young Scholars (No.62025603) and National Natural Science Foundation of China (No. 62072386), supported by Yunnan Provincial Major S&T Special Plan Project (No. 202402AD080001), Henan Key R&D Project (No. 231111212000), Henan Center for Outstanding Overseas Scientists (No. GZS2022011), Open Foundation of Henan Key Lab of General Aviation Technology (No. ZHKF-230212), Key Lab of Oracle Information Processing of MOE (No. OIP2024E002). We also thank Digital Culture Lab of Tencent SSV for their invaluable contribution during our research." + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.945, + 0.519, + 0.958 + ], + "angle": 0, + "content": "22623" + } + ], + [ + { + "type": "title", + "bbox": [ + 0.093, + 0.09, + 0.188, + 0.106 + ], + "angle": 0, + "content": "References" + }, + { + "type": "ref_text", + "bbox": [ + 0.101, + 0.115, + 0.484, + 0.156 + ], + "angle": 0, + "content": "[1] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint arXiv:1607.06450, 2016. 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.101, + 0.159, + 0.484, + 0.242 + ], + "angle": 0, + "content": "[2] Ankan Kumar Bhunia, Salman Khan, Hisham Cholakkal, Rao Muhammad Anwer, Jorma Laaksonen, Mubarak Shah, and Fahad Shahbaz Khan. Person image synthesis via denoising diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5968-5976, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.245, + 0.483, + 0.285 + ], + "angle": 0, + "content": "[3] Mikołaj Binkowski, Danica J Sutherland, Michael Arbel, and Arthur Gretton. Demystifying mmd gans. arXiv preprint arXiv:1801.01401, 2018. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.288, + 0.483, + 0.343 + ], + "angle": 0, + "content": "[4] Shidong Cao, Wenhao Chai, Shengyu Hao, Yanting Zhang, Hangyue Chen, and Gaoang Wang. Diffashion: Reference-based fashion design with structure-aware transfer by diffusion models. IEEE Transactions on Multimedia, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.345, + 0.483, + 0.4 + ], + "angle": 0, + "content": "[5] Mengting Chen, Xi Chen, Zhonghua Zhai, Chen Ju, Xuewen Hong, Jinsong Lan, and Shuai Xiao. Wear-any-way: Manipulable virtual try-on via sparse correspondence alignment. arXiv preprint arXiv:2403.12965, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.403, + 0.483, + 0.471 + ], + "angle": 0, + "content": "[6] Seunghwan Choi, Sunghyun Park, Minsoo Lee, and Jaegul Choo. Viton-hd: High-resolution virtual try-on via misalignment-aware normalization. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14131-14140, 2021. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.474, + 0.483, + 0.514 + ], + "angle": 0, + "content": "[7] Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for virtual try-on. ECCV, 2024. 3, 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.517, + 0.483, + 0.585 + ], + "angle": 0, + "content": "[8] Zheng Chong, Xiao Dong, Haoxiang Li, Shiyue Zhang, Wenqing Zhang, Xujie Zhang, Hanqing Zhao, and Xiaodan Liang. Catvton: Concatenation is all you need for virtual try-on with diffusion models. arXiv preprint arXiv:2407.15886, 2024. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.102, + 0.589, + 0.483, + 0.657 + ], + "angle": 0, + "content": "[9] Morelli Davide, Fincato Matteo, Cornia Marcella, Landi Federico, Cesari Fabio, and Cucchiara Rita. Dress code: High-resolution multi-category virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.66, + 0.483, + 0.728 + ], + "angle": 0, + "content": "[10] Haoye Dong, Xiaodan Liang, Xiaohui Shen, Bochao Wang, Hanjiang Lai, Jia Zhu, Zhiting Hu, and Jian Yin. Towards multi-pose guided virtual try-on network. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9026-9035, 2019." + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.731, + 0.483, + 0.799 + ], + "angle": 0, + "content": "[11] Chongjian Ge, Yibing Song, Yuying Ge, Han Yang, Wei Liu, and Ping Luo. Disentangled cycle consistency for highly-realistic virtual try-on. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16928-16937, 2021. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.802, + 0.483, + 0.87 + ], + "angle": 0, + "content": "[12] Yuying Ge, Yibing Song, Ruimao Zhang, Chongjian Ge, Wei Liu, and Ping Luo. Parser-free virtual try-on via distilling appearance flows. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8485-8493, 2021. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.094, + 0.873, + 0.483, + 0.901 + ], + "angle": 0, + "content": "[13] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and" + }, + { + "type": "list", + "bbox": [ + 0.094, + 0.115, + 0.484, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "ref_text", + "bbox": [ + 0.548, + 0.093, + 0.905, + 0.12 + ], + "angle": 0, + "content": "Yoshua Bengio. Generative adversarial networks. Communications of the ACM, 63(11):139-144, 2020. 1, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.123, + 0.905, + 0.191 + ], + "angle": 0, + "content": "[14] Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang. Taming the power of diffusion models for high-quality virtual try-on with appearance flow. In Proceedings of the 31st ACM International Conference on Multimedia, pages 7599-7607, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.193, + 0.905, + 0.248 + ], + "angle": 0, + "content": "[15] Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S Davis. Viton: An image-based virtual try-on network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7543-7552, 2018. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.25, + 0.905, + 0.305 + ], + "angle": 0, + "content": "[16] Xintong Han, Xiaojun Hu, Weilin Huang, and Matthew R Scott. Clothflow: A flow-based model for clothed person generation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 10471-10480, 2019." + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.307, + 0.905, + 0.362 + ], + "angle": 0, + "content": "[17] Sen He, Yi-Zhe Song, and Tao Xiang. Style-based global appearance flow for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3470-3479, 2022. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.364, + 0.905, + 0.432 + ], + "angle": 0, + "content": "[18] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems, 30, 2017. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.435, + 0.905, + 0.503 + ], + "angle": 0, + "content": "[19] Xiaobin Hu, Xu Peng, Donghao Luo, Xiaozhong Ji, Jinlong Peng, Zhengkai Jiang, Jiangning Zhang, Taisong Jin, Chengjie Wang, and Rongrong Ji. Diffumatting: Synthesizing arbitrary objects with matting-level annotation. arXiv preprint arXiv:2403.06168, 2024. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.505, + 0.905, + 0.574 + ], + "angle": 0, + "content": "[20] Catalin Ionescu, Dragos Papava, Vlad Olaru, and Cristian Sminchisescu. Human3. 6m: Large scale datasets and predictive methods for 3d human sensing in natural environments. IEEE transactions on pattern analysis and machine intelligence, 36(7):1325-1339, 2013. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.577, + 0.905, + 0.646 + ], + "angle": 0, + "content": "[21] Thibaut Issenhuth, Jérémie Mary, and Clément Calauzenes. Do not mask what you do not need to mask: a parser-free virtual try-on. In Computer Vision-ECCV 2020: 16th European Conference, Glasgow, UK, August 23-28, 2020, Proceedings, Part XX 16, pages 619-635. Springer, 2020. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.648, + 0.905, + 0.716 + ], + "angle": 0, + "content": "[22] Johanna Karras, Aleksander Holynski, Ting-Chun Wang, and Ira Kemelmacher-Shlizerman. Dreampose: Fashion image-to-video synthesis via stable diffusion. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 22623-22633. IEEE, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.718, + 0.905, + 0.787 + ], + "angle": 0, + "content": "[23] Bahjat Kawar, Shiran Zada, Oran Lang, Omer Tov, Huiwen Chang, Tali Dekel, Inbar Mosseri, and Michal Irani. Imagic: Text-based real image editing with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6007-6017, 2023. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.789, + 0.905, + 0.858 + ], + "angle": 0, + "content": "[24] Jeongho Kim, Guojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8176-8185, 2024. 3, 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.518, + 0.86, + 0.905, + 0.901 + ], + "angle": 0, + "content": "[25] Nupur Kumari, Bingliang Zhang, Richard Zhang, Eli Shechtman, and Jun-Yan Zhu. Multi-concept customization of text-to-image diffusion. In Proceedings of the IEEE/CVF" + }, + { + "type": "list", + "bbox": [ + 0.518, + 0.093, + 0.905, + 0.901 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.946, + 0.519, + 0.957 + ], + "angle": 0, + "content": "22624" + } + ], + [ + { + "type": "ref_text", + "bbox": [ + 0.126, + 0.092, + 0.482, + 0.121 + ], + "angle": 0, + "content": "Conference on Computer Vision and Pattern Recognition, pages 1931-1941, 2023. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.124, + 0.483, + 0.193 + ], + "angle": 0, + "content": "[26] Sangyun Lee, Gyojung Gu, Sunghyun Park, Seunghwan Choi, and Jaegul Choo. High-resolution virtual try-on with misalignment and occlusion-handled conditions. In European Conference on Computer Vision, pages 204-219. Springer, 2022. 1, 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.195, + 0.483, + 0.25 + ], + "angle": 0, + "content": "[27] Yuhan Li, Hao Zhou, Wenxiang Shang, Ran Lin, Xuanhong Chen, and Bingbing Ni. Anyfit: Controllable virtual try-on for any combination of attire across any scenario. arXiv preprint arXiv:2405.18172, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.253, + 0.482, + 0.281 + ], + "angle": 0, + "content": "[28] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.284, + 0.482, + 0.338 + ], + "angle": 0, + "content": "[29] Wenquan Lu, Yufei Xu, Jing Zhang, Chaoyue Wang, and Dacheng Tao. Handrefiner: Refining malformed hands in generated images by diffusion-based conditional inpainting. In ACM Multimedia 2024, 2023. 4, 7" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.342, + 0.482, + 0.411 + ], + "angle": 0, + "content": "[30] Yifang Men, Yiming Mao, Yuning Jiang, Wei-Ying Ma, and Zhouhui Lian. Controllable person image synthesis with attribute-decomposed gan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5084-5093, 2020. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.414, + 0.483, + 0.483 + ], + "angle": 0, + "content": "[31] Davide Morelli, Matteo Fincato, Marcella Cornia, Federico Landi, Fabio Cesari, and Rita Cucchiara. Dress code: High-resolution multi-category virtual try-on. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2231-2235, 2022. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.486, + 0.482, + 0.555 + ], + "angle": 0, + "content": "[32] Davide Morelli, Alberto Baldrati, Giuseppe Cartella, Marcelia Cornia, Marco Bertini, and Rita Cucchiara. Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on. In Proceedings of the 31st ACM International Conference on Multimedia, pages 8580-8589, 2023. 3, 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.558, + 0.482, + 0.627 + ], + "angle": 0, + "content": "[33] Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4296–4304, 2024. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.63, + 0.482, + 0.699 + ], + "angle": 0, + "content": "[34] Maxime Oquab, Timothee Darcet, Theo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.702, + 0.483, + 0.77 + ], + "angle": 0, + "content": "[35] Georgios Pavlakos, Dandan Shan, Ilija Radosavovic, Angjoo Kanazawa, David Fouhey, and Jitendra Malik. Reconstructing hands in 3d with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9826-9836, 2024. 4, 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.774, + 0.482, + 0.829 + ], + "angle": 0, + "content": "[36] Anton Pelykh, Ozge Mercanoglu Sincan, and Richard Bowden. Giving a hand to diffusion models: a two-stage approach to improving conditional human image generation. arXiv preprint arXiv:2403.10731, 2024. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.832, + 0.482, + 0.9 + ], + "angle": 0, + "content": "[37] Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952, 2023. 4" + }, + { + "type": "list", + "bbox": [ + 0.093, + 0.092, + 0.483, + 0.9 + ], + "angle": 0, + "content": null + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.093, + 0.905, + 0.148 + ], + "angle": 0, + "content": "[38] Sergey Prokudin, Christoph Lassner, and Javier Romero. Efficient learning on point clouds with basis point sets. In Proceedings of the IEEE/CVF international conference on computer vision, pages 4332-4341, 2019. 5" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.151, + 0.907, + 0.219 + ], + "angle": 0, + "content": "[39] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684-10695, 2022. 1, 4, 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.221, + 0.905, + 0.303 + ], + "angle": 0, + "content": "[40] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22500-22510, 2023. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.306, + 0.905, + 0.373 + ], + "angle": 0, + "content": "[41] Chitwan Sahara, William Chan, Huiwen Chang, Chris Lee, Jonathan Ho, Tim Salimans, David Fleet, and Mohammad Norouzi. Palette: Image-to-image diffusion models. In ACM SIGGRAPH 2022 conference proceedings, pages 1-10, 2022. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.376, + 0.905, + 0.432 + ], + "angle": 0, + "content": "[42] Dvir Samuel, Rami Ben-Ari, Simon Raviv, Nir Darshan, and Gal Chechik. Generating images of rare concepts using pretrained diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4695-4703, 2024. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.434, + 0.905, + 0.502 + ], + "angle": 0, + "content": "[43] Ke Sun, Jian Cao, Qi Wang, Linrui Tian, Xindi Zhang, Lian Zhuo, Bang Zhang, Liefeng Bo, Wenbo Zhou, Weiming Zhang, et al. Outfitanyone: Ultra-high quality virtual try-on for any clothing and any person. arXiv preprint arXiv:2407.16224, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.505, + 0.905, + 0.56 + ], + "angle": 0, + "content": "[44] Chengrui Wang, Pengfei Liu, Min Zhou, Ming Zeng, Xubin Li, Tiezheng Ge, et al. Rhand: Refining malformed hands for generated images with decoupled structure and style guidance. arXiv preprint arXiv:2404.13984, 2024. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.562, + 0.905, + 0.617 + ], + "angle": 0, + "content": "[45] Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600-612, 2004. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.619, + 0.905, + 0.672 + ], + "angle": 0, + "content": "[46] Yuxin Wu, Alexander Kirillov, Francisco Massa, Wan-Yen Lo, and Ross Girshick. Detectron2. https://github.com/facebookresearch/detectron2, 2019. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.676, + 0.907, + 0.759 + ], + "angle": 0, + "content": "[47] Zhenyu Xie, Zaiyu Huang, Xin Dong, Fuwei Zhao, Haoye Dong, Xijin Zhang, Feida Zhu, and Xiaodan Liang. Gqvton: Towards general purpose virtual try-on via collaborative local-flow global-parsing learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 23550-23559, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.761, + 0.905, + 0.815 + ], + "angle": 0, + "content": "[48] Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen. Oot-diffusion: Outfitting fusion based latent diffusion for controllable virtual try-on. arXiv preprint arXiv:2403.01779, 2024.6" + }, + { + "type": "ref_text", + "bbox": [ + 0.517, + 0.818, + 0.905, + 0.899 + ], + "angle": 0, + "content": "[49] Binxin Yang, Shuyang Gu, Bo Zhang, Ting Zhang, Xuejin Chen, Xiaoyan Sun, Dong Chen, and Fang Wen. Paint by example: Exemplar-based image editing with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18381-18391, 2023. 3" + }, + { + "type": "list", + "bbox": [ + 0.517, + 0.093, + 0.907, + 0.899 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.946, + 0.518, + 0.957 + ], + "angle": 0, + "content": "22625" + } + ], + [ + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.093, + 0.482, + 0.161 + ], + "angle": 0, + "content": "[50] Han Yang, Ruimao Zhang, Xiaobao Guo, Wei Liu, Wangmeng Zuo, and Ping Luo. Towards photo-realistic virtual try-on by adaptively generating-preserving image content. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7850-7859, 2020. 1" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.164, + 0.482, + 0.218 + ], + "angle": 0, + "content": "[51] Zhijing Yang, Junyang Chen, Yukai Shi, Hao Li, Tianshui Chen, and Liang Lin. Occlumix: Towards de-occlusion virtual try-on by semantically-guided mixup. IEEE Transactions on Multimedia, 25:1477–1488, 2023. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.22, + 0.482, + 0.275 + ], + "angle": 0, + "content": "[52] Zhendong Yang, Ailing Zeng, Chun Yuan, and Yu Li. Effective whole-body pose estimation with two-stages distillation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4210-4220, 2023. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.277, + 0.482, + 0.33 + ], + "angle": 0, + "content": "[53] Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv preprint arXiv:2308.06721, 2023.4" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.333, + 0.482, + 0.402 + ], + "angle": 0, + "content": "[54] Jianhao Zeng, Dan Song, Weizhi Nie, Hongshuo Tian, Tongtong Wang, and An-An Liu. Cat-dm: Controllable accelerated virtual try-on with diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8372-8382, 2024. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.404, + 0.482, + 0.458 + ], + "angle": 0, + "content": "[55] Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, Jun Zhu, Lionel M Ni, and Heung-Yeung Shum. Dino: Detr with improved denoising anchor boxes for end-to-end object detection. ICLR, 2023. 2" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.461, + 0.482, + 0.502 + ], + "angle": 0, + "content": "[56] Haozhuo Zhang, Bin Zhu, Yu Cao, and Yanbin Hao. Hand1000: Generating realistic hands from text with only 1,000 images. arXiv preprint arXiv:2408.15461, 2024. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.504, + 0.482, + 0.558 + ], + "angle": 0, + "content": "[57] Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3836-3847, 2023. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.56, + 0.482, + 0.628 + ], + "angle": 0, + "content": "[58] Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 586-595, 2018. 6" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.631, + 0.482, + 0.685 + ], + "angle": 0, + "content": "[59] Xujie Zhang, Ente Lin, Xiu Li, Yuxuan Luo, Michael Kampffmeyer, Xin Dong, and Xiaodan Liang. Mmtryon: Multi-modal multi-reference control for high-quality fashion generation. arXiv preprint arXiv:2405.00448, 2024. 3" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.687, + 0.482, + 0.755 + ], + "angle": 0, + "content": "[60] Shihao Zhao, Dongdong Chen, Yen-Chun Chen, Jianmin Bao, Shaozhe Hao, Lu Yuan, and Kwan-Yee K Wong. Uni-controlnet: All-in-one control to text-to-image diffusion models. Advances in Neural Information Processing Systems, 36, 2024. 4" + }, + { + "type": "ref_text", + "bbox": [ + 0.093, + 0.758, + 0.482, + 0.839 + ], + "angle": 0, + "content": "[61] Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Sahara, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two unets. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4606-4615, 2023. 1, 2, 3" + }, + { + "type": "list", + "bbox": [ + 0.093, + 0.093, + 0.482, + 0.839 + ], + "angle": 0, + "content": null + }, + { + "type": "page_number", + "bbox": [ + 0.48, + 0.946, + 0.519, + 0.957 + ], + "angle": 0, + "content": "22626" + } + ] +] \ No newline at end of file diff --git a/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/00462254-bb4b-4c24-9d0a-9a8f2054757a_origin.pdf b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/00462254-bb4b-4c24-9d0a-9a8f2054757a_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..349e6f198a06555c0b614f5fa161b933e3c56fe4 --- /dev/null +++ b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/00462254-bb4b-4c24-9d0a-9a8f2054757a_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:53f9d0e74a3c92be4664106578ce447a14be1065660c7007c9ce14b5ae087aff +size 7436649 diff --git a/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/full.md b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/full.md new file mode 100644 index 0000000000000000000000000000000000000000..89a788fa8c007ba2119293efa5a71ecfa93190d4 --- /dev/null +++ b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/full.md @@ -0,0 +1,257 @@ +# VTON-HandFit: Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding + +Yujie Liang $^{1,2*}$ , Xiaobin Hu $^{3*}$ , Boyuan Jiang $^{3}$ , Donghao Luo $^{3\dagger}$ , Xu Peng $^{3}$ , Kai Wu $^{3}$ , Chengming Xu $^{3}$ , Wenhui Han $^{3}$ , Taisong Jin $^{1,2\dagger}$ , Chengjie Wang $^{3}$ , Rongrong Ji $^{1,2}$ $^{1}$ Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, China. + $^{2}$ School of Informatics, Xiamen University, China. $^{3}$ Tencent Youtu Lab. + +liangyujie@stu.xmu.edu.cn,{xiaobinhu,byronjiang,michaelluo,ppxupeng, chengmingxu,jasoncjwang}@tencent.com,wukaiwork,outlook.com, whhan18@gmail.com,{jintaisong,rrji}@xmu.edu.cn + +# Abstract + +Although diffusion-based image virtual try-on has made considerable progress, emerging approaches still struggle to effectively address the issue of hand occlusion (i.e., clothing regions occluded by the hand part), leading to a notable degradation of the try-on performance. To tackle this issue widely existing in real-world scenarios, we propose VTON-HandFit, leveraging the power of hand priors to reconstruct the appearance and structure for hand occlusion cases. Firstly, we tailor a Hand-Pose Aggregation Net using the ControlNet-based structure explicitly and adaptively encoding the global hand and pose priors. Besides, to fully exploit the hand-related structure and appearance information, we propose Hand-feature Disentanglement Embedding module to disentangle the hand priors into the hand structure-parametric and visual-appearance features, and customize a masked cross attention for further decoupled feature embedding. Lastly, we customize a hand-canny constraint loss to better learn the structure edge knowledge from the hand template of model image. VTON-HandFit outperforms the baselines in qualitative and quantitative evaluations on the public dataset and our self-collected hand-occlusion Handfit-3K dataset particularly for the arbitrary hand pose occlusion cases in real-world scenarios. Our project page is at: https://vton-handfit.github.io. + +# 1. Introduction + +Image-based virtual try-on (VTON) [26] is a widely adopted and highly promising image synthesis technology in the e-commerce industry. Its primary goal is to enhance + +the shopping experience for consumers and minimize advertising expenses for clothing merchants. The VTON task involves generating an image of a human model wearing a specific garment. Over the past few years, countless researchers [9-11, 50] have dedicated significant efforts to achieve more realistic and precise virtual try-on results. + +Most recent studies on virtual try-on predominantly utilize generative adversarial networks (GANs) [13] or latent diffusion models (LDMs) [39] for image synthesis. Traditional GAN-based approaches [15-17] often struggle with accurately generating garment folds, realistic lighting and shadows, and lifelike human body representations. Consequently, more recent research has shifted towards LDM-based methods [61], which effectively improve the authenticity of clothed images. However, these researches focus on generating more realistic and natural try-on images while keeping the detailed features of garments. The hands of try-on models are usually strictly on the sides and garments are not occluded by the hands. Such a strict constraint hinders the virtual try-on paving the way toward real-world applications. For instance, as a highly flexible editing task with a wide range of degrees of freedom, digital humans usually meet the occasions that the garments are occluded by the hands, which severely deteriorates the performance of virtual try-on. + +Motivated by the above challenge, basically, there mainly exist two solutions to tackle this arbitrary hand problem: 1). perfect hand parsing model to accurately segment the tiny area among fingers. 2). involving the hand-reconstruction into the try-on processing. As shown in Fig. 1, CatVTON[8] use the parsing model to keep the hand part, leading to unavoidably the remaining artifacts and backgrounds from the model image. The existing parsing models fail to accurately segment the small + +![](images/ffb33d5ded6172475685713deedb9529d89316456ef43ff78a83006a4782f928.jpg) +Figure 1. Comparison of different models on virtual try-on with hand occlusion. The leftmost three images are reference images, showing the target clothing, the model, and a close-up of the model's hand. Our model excels in preserving hand details and achieving realistic clothing transfer, as highlighted in the red boxes. CatVTON utilizes a parsing model to retain the hand segment, resulting in the inevitable persistence of residual artifacts and backgrounds from the model image. IDM-VTON* indicates post-processing using HandRefiner, which fails to address missing hands and mistakenly alters correctly depicted ones. + +and fragmented finger areas, thus we embrace the latter strategy to involve the hand-reconstruction by disentangling the hand into the appearance and structure knowledge processing in the try-on stage. Then, we propose a novel diffusion-based virtual try-on method, termed as VTON-HandFit, to generate realistic and natural images for arbitrary hand position or hand-occlusion cases with the aid of hand 3D structure-parametric and visual-appearance priors. To embed the pose- and hand-related knowledge, we build a Hand-Pose Net to exploit the body and hand semantic and structure information by getting the body (e.g., DWPose, DensPose) and 3D hand HaMeR-based [61] reconstruction information. Secondly, considering the over-small character of hand that is easily ignored after down-sampling operation, we use the DINO [55] to dig the local appearance and texture knowledge of hand as the hand-visual embedding features. For the structure of hand, we adopt these structure-parametric priors (e.g., 3D MANO-related parameters, handtype and hand-box coordinates) as hand-structure embedding features. After getting structure-parametric and visual-appearance features, we aim to enhance the model's ability to respectively learn hand-related patterns. To achieve this, we inject the hand-structure embedding into the Hand-Pose Net with a primary focus on enhancing the latent features related to hand structure, and the hand-visual embedding into the mainstream U-Net specifically aiming to enhance texture-related features, respec + +tively. The disentanglement learning of structure and visual appearance enhances the network's ability to efficiently and respectively acquire hand-related knowledge. To prevent the interference of hand-related priors with garment features, we impose a constraint that restricts the hand-related features to operate solely within the hand mask area of the cross-attention mechanism. Finally, we tailor a handcanny constraint loss to more effectively acquire the structural edge knowledge from the hand template of the model image. Our contributions are summarized as follows: + +- We propose VTON-HandFit, a novel virtual try-on architecture considering to solve the Hand pose occlusion problem popularly existing in real-world scenarios. +- We propose a Hand-Pose Aggregation Net to encode the hand and pose structure priors and also adaptively to adjust the weight of each component. Meanwhile, hand-canny constraint loss is designed to learn the hand structural-edge knowledge of model images. +- We propose a novel hand-feature disentanglement embedding module disentangling the hand knowledge into the structure-parametric priors (e.g., MANO parameters) and visual-appearance priors (e.g., color). A mask cross-attention mechanism is proposed to constrain the hand-related features only working on the hand mask area to avoid the garment feature disturbance. +- We self-collect Handfit-3K test dataset from online retail + +![](images/74037f3537d914e6a0e9804d1cc30bb05894de5764f3a14727b1395c8c65a240.jpg) +Figure 2. An overview of our VTON-HandFit Network. The network consists of two main components: Hand-feature Disentanglement Embedding and Hand-Pose Aggregation Net. The Hand-feature Disentanglement Embedding module uses the HaMer model to extract hand priors, including hand type $T_{h}$ , 3D vertices $V_{h}$ , spatial joint locations $J_{2d}$ , and joint rotation matrices $\theta_{h}$ . These features are processed by the Hand-Struct processor to derive structural features $c_{struc}$ . Simultaneously, hand images cropped using bounding boxes are processed by DINOv2 and the Hand-Appear processor to obtain visual features $c_{appear}$ . The structural and visual features are integrated using mask cross attention. The Hand-Pose Aggregation Net module controls body and hand poses by aggregating DWpose, Densepose, and hand depth maps. + +sites to evaluate hand occlusion performance. Extensive qualitative and quantitative evaluations demonstrate our superiority over state-of-the-art virtual try-on models especially for hand pose occlusion cases, paving the road towards more complicated virtual try-on in real-world. + +# 2. Related Works + +Image-based virtual try-on. In the context of image-based virtual try-on, the objective is to generate a realistic representation of a target person wearing a specific garment, given a pair of corresponding images. A series of studies [26, 30, 47, 51] have been conducted using Generative Adversarial Networks (GANs) [13]. These studies typically involve the initial deformation of the garment to match the body shape of the person, followed by the utilization of a generator to superimpose the deformed garment onto the person image. Despite efforts made to minimize the discrepancy between the warped garment and the person [12, 21, 26], these approaches often lack the ability to generalize well to diverse person images, particularly those with complex backgrounds or intricate poses. There exists + +GAN-based try-on method [26] considers occluded problems and designs the warping operation to avoid the garment warping misalignment, which is totally different from diffusion-based algorithms as an inpainting problem. + +The remarkable generative capabilities of diffusion models have sparked interest in incorporating them into fashion synthesis, particularly in tasks like visual try-on [2, 4, 5, 22, 27, 43, 59]. TryOnDiffusion [61] employs ParallelUNets to preserve garment details and warp the garment for the try-on process. Subsequent studies have treated virtual try-on as an exemplar-based image inpainting problem [49]. These studies involved fine-tuning the inpainting diffusion models using virtual try-on datasets [24, 32] to generate virtual try-on images of superior quality. For instance, such as LADI-VTON [32] and DCI-VTON [14] have emerged, treating clothing as pseudo-words or employing warping networks to integrate garments into pre-trained diffusion models. IDM-VTON [7] incorporates sophisticated attention modules to encode the high-level semantics of the garment and extract low-level features, thereby preserving fine-grained details. While these diffusion methods address issues related to the nature and realism of synthesized images, + +they struggle to meet hard occasions of the hand-based occlusion which very widely exists in the real-world virtual try-on scenarios. + +Conditional control of diffusion models. In recent years, latent diffusion models [39] have demonstrated significant success in tasks such as text-to-image generation [19, 25, 37, 40] and image-to-image generation [23, 41]. To achieve more controllable generated results, several studies have introduced conditional control to diffusion models [53, 60]. ControlNet [57] adds spatial conditioning controls (e.g.,anny edges, Hough lines, user scribbles, human key points) to pretrained diffusion models via zero convolutions manner. T2I-Adapter [33] aligns the internal knowledge within T2I models with external control signals, allowing it to be trained based on different conditions. This approach enables the generation of results with rich control and editing effects, particularly in terms of color and structure. In our paper, we focus on the hard occasions of hand-based occlusion in image-based VTON task, and tailor a Hand-Pose Aggregation Net to control the body pose and hand structure. Furthermore, the disengagement knowledge of hand structure and appearance embedding is studied and then these embeddings are respectively injected into the attention mechanism, acting as a mask-based regional control. + +Reliable human hand generation. Previous studies [42] have acknowledged the problem of misshaped hands that are generated by existing text-to-image diffusion models. Controlling the hand is considerably more challenging compared to controlling the body pose, primarily due to the relative size of the main subject. Besides, diffusion models exhibit significantly improved capability in generating plausible whole-body poses. This difficulty arises from the intricate nature of learning the physical structure and pose of hands from training images, which encompasses extensive deformations and occlusions. Handrefiner [29] and RHanDS [44] are post-processing solutions adopting a hand mesh reconstruction model to rectify malformed hands. Hand-to-Diffusion [36] uses a two-stage approach, first generating hand effects and then producing the entire image. Hand1000 [56] employs a three-stage training process that integrates hand gesture recognition and optimized text embeddings. VTON-HandFit involves the hand corrections into the generation process not as a post-processing and disentangles the hand priors into the structure and appearance feature embedding for virtual try-on arbitrary hand pose in the real-world. + +# 3. Methodology + +Data collection of Handfit-3K. Considering that the hand of the current main virtual try-on dataset is separate from the body, there rarely exists a great amount of virtual try-on data for arbitrary hand pose and hard hand-occlusion cases. To meet this urgent need in real-world, we aim to + +construct a dataset of complex hand poses that occlude the garment. Specifically, we collect an additional dataset of 26,000 images sourced from online retail sites. These images underwent processing using a parsing model and a hand bounding box via HaMeR model [35] to identify instances of occlusion. We devote significant manual effort to inspecting the quality of the data, removing any erroneous or flawed entries, thereby ensuring the high quality of the virtual try-on dataset. As a result, we obtained a specialized hand occlusion test set comprising 3,620 images, termed as Handfit-3K. This dataset facilitates a targeted evaluation of our method's ability to effectively handle occlusions. + +# 3.1. Preliminary + +Latent Diffusion Models. Our VTON-HandFit is an extension of stable diffusion model, which is a probabilistic model specifically designed to estimate a data distribution $\mathfrak{p}(\mathbf{x})$ by iteratively reducing noise in a normally distributed variable. Given an input image $x_0$ , the noise estimation process is: + +$$ +\mathcal {L} _ {\text {n o i s e}} = \mathbb {E} _ {z \sim \mathcal {E} (x), C, \epsilon \sim \mathcal {N} (0, 1), t} \left[ | | \epsilon - \epsilon_ {\theta} (z _ {t}, t, C) | | _ {2} ^ {2} \right], \tag {1} +$$ + +where a Variational AutoEncoder $\mathcal{E}$ compress input image $x$ into low-dimension latent space $z$ . A conditional U-Net denoiser $\epsilon_{\theta}$ aims to estimate noise $\theta$ in latent space with the aid of timestep $t$ , $t$ -th noisy latent representation $z_{t}$ , and other text-prompt conditions $C$ extracted by text-encoder. + +# 3.2.VTON-HandFit + +Overview. An overview of our VTON-HandFit is illustrated in Fig. 2. Given a target person image $x \in \mathbb{R}^{H \times W \times 3}$ and an input garment image $g \in \mathbb{R}^{H \times W \times 3}$ , VTON-HandFit aims to generate a realistic outfitted image $x_{g} \in \mathbb{R}^{H \times W \times 3}$ . The garment-agnostic person representation $x_{m}$ is masked to erase the garment information by the combination of HumanParsing and HumanPose. Thus, this virtual try-on task is transformed into an example-based inpainting problem to recover the missing mask part. For the input of TryonNet, we simply concatenate the noise image $(z_{t})$ and latent garment-agnostic features $\mathcal{E}(x_{m})$ to avoid the other pose-relevant information disturbances. In Hand-Pose Aggregation Net, we also aggregate the pose-related knowledge including the DWPose, DensePose and 3D reconstruction depth map derived by the HaMeR mesh model. These pose-related priors are detached with TryonNet to explicitly be learned in Hand-Pose Net. DensePose provides more bodysemantic information, DWPose contains abundant skeleton structure, and 3D-hand-derived depth maps demonstrate the pixel-level global-aligned hand position and structure knowledge. These modules complement each other by mutually providing information, and working together to constrain the pose and hand generation. In hand-feature disentanglement embedding, we use the HaMeR model for 3D + +hand reconstruction and get the structure-parametric priors. These features are processed by the Hand-Struct processor to derive structural features $c_{struc}$ . Simultaneously, cropped hand images are processed by DINOv2 and the Hand-Appear processor to obtain visual features $c_{appear}$ . Besides, GarmentNet aims to extract the garment features into the TryonNet mainly by: 1). encoding the garment image into the latent space $\mathcal{E}_g$ , 2). using the CLIP to get the image token features for injection. + +Hand-Pose Aggregation Net. To aggregate the hand and pose-relevant priors learning, we tailor a Hand-Pose Net to constrain the semantic and skeleton pose, and global pixel-level hand knowledge. Specifically, dwpose, densepose, and render blocks are designed based on the same zero-multilayer structure with different weights to respectively extract the corresponding features. Zero-multilayer structure consists of seven stacking of conv and SiLU activate function with the zero convolution layer at the last layer. + +$$ +\mathcal {F} _ {h p} = \mathcal {F} _ {d w} + \mathcal {F} _ {d p} + \mathcal {F} _ {r h} \times w _ {\text {h a n d}}, \tag {2} +$$ + +where $\mathcal{F}_{dw}$ , $\mathcal{F}_{dp}$ , $\mathcal{F}_{rh}$ are the DWPose feature, Densepose feature, and hand feature after the Zero-multilayer structure, and then the summation is implemented on the pixel-level with the adjustable weight $w_{\mathrm{hand}}$ of hand feature $\mathcal{F}_{rh}$ . To mitigate the impact of detrimental noise on the hidden states of the neural network layers in the trainable Hand-Pose Net during the initial stages of training, we also adopt the zero convolution learning paradigm of ControlNet. + +Hand structural embedding. The HaMeR [35] model is utilized to reconstruct various MANO parameters and other hand-related information from 2D images, including hand mask $M_h \in \mathbb{R}^{768 \times 1024}$ , hand types $T_h \in \mathbb{R}^{n \times 1}$ , hand 3D vertices $V_h \in \mathbb{R}^{n \times 778 \times 3}$ , spatial joint locations $J_{2d} \in \mathbb{R}^{n \times 21 \times 2}$ , and joint rotation matrices $\theta \in \mathbb{R}^{n \times 16 \times 3 \times 3}$ where $n$ represents the number of hands. To accommodate variable numbers of hands as input, we pad the hand parameters to a fixed length $n = 2$ by adding padding at the end. For hand type $T_h$ , we use 0 to denote left hands, 1 for right hands, and -1 as a filler for images without hands. Similar padding is applied to other parameters to ensure alignment and correspondence across all dimensions. We apply Basis Point Set (BPS) [38] for dimensionality reduction of hand vertices and encode joint rotations as 6D vectors. And then, we import theses information as well as joint rotations and hand types into the hand-structure processor. The structure of hand-structure processor consists of two parallel blocks: three stacked layers of linear and ReLU activation $(L_r)$ and two stacked layers of linear and layer normalization $(L_n)$ . These streamlined hand vertices and joint locations are then fed into $L_r$ . Concurrently, joint rotations and hand types are transformed via $L_n$ . The outputs from these processes are aggregated to form the Hand Structural Embedding $c_{struc}$ , encapsulating essential geometric and 3D information for + +precise hand representation. + +$$ +c _ {s t r u c} = \left[ L _ {r} \left(\boldsymbol {V} _ {\boldsymbol {h}}\right), L _ {r} \left(\boldsymbol {J} _ {\boldsymbol {2 d}}\right), L _ {n} \left(\boldsymbol {T} _ {\boldsymbol {h}}\right), L _ {n} \left(\boldsymbol {\theta} _ {\boldsymbol {h}}\right) \right]. \tag {3} +$$ + +Hand appearance embedding. To improve hand visual detail, we extract features $\mathcal{F} \in \mathbb{R}^{n \times 1536}$ from the hand Bounding Box regions using DINOv2 [34]. The Hand-Appear processor (HA) uses a linear layer followed by Layer Normalization [1], culminating in the Hand Appearance Embedding $c_{\text{appear}}$ . This process ensures the hand's visual fidelity and detail are meticulously preserved in our representation. + +$$ +c _ {\text {a p p e a r}} = \mathrm {H A} (\mathcal {F}). \tag {4} +$$ + +Mask-based regional control. Our method introduces a mask-based regional control to enhance hand detail generation and maintain quality in virtual try-on. Utilizing the hand mask $M_h$ , we modulate Cross-Attention queries, seamlessly integrating Hand Structural Embedding within Hand-Pose Net's Cross-Attention module. Concurrently, Hand Appearance Embedding is adeptly fused into the Try-onNet architecture through mask cross-attention, ensuring precise detail preservation. The mask cross-attention can be represented as: + +$$ +\begin{array}{l} Z = \operatorname {A t t e n t i o n} (Q, K, V, M _ {h}) \\ = \operatorname {S o f t m a x} \left(\frac {\left(Q K ^ {\top}\right) \odot M _ {h}}{\sqrt {d}}\right) V, \tag {5} \\ \end{array} +$$ + +where $Q, K$ and $V$ are the query, key, and value matrices of the attention operation for hand appearance embedding cross-attention, $M_h$ represents the hand mask. + +Hand canny shape constraint. During the training process, $\mathcal{L}_{hand}$ is implemented to refine the clarity and precision of hand edges. Specifically, for a small $R_{t} < T$ , we employ one-step denoising process prior to sampling $x_0$ for the direct prediction of $z_0$ from $z_{t}$ and the model-predicted noise, using Variational Autoencoder for decoder. + +$$ +z _ {0} = \frac {z _ {t} - \sqrt {1 - \bar {\alpha} _ {t}} \epsilon_ {\theta}}{\sqrt {\bar {\alpha} _ {t}}}, \quad t \leq R _ {t}. \tag {6} +$$ + +Hand regions are extracted from the generated images using bounding box cropping and are processed with the Canny edge detection operator. These regions are then compared to their corresponding areas in the Ground Truth images using Mean Squared Error (MSE) loss. This loss function is specifically designed to sharpen the model's focus on improving edge details and clarity in the hand regions of the generated images. + +# 4. Experiment + +Baselines. To comprehensively evaluate our approach, we selected five state-of-the-art diffusion-based methods in the + +![](images/87c2702bedbfbc89cfbca50f4f4eb62cfa75b3a211cc8e5d581619b6cc4c3ee6.jpg) + +![](images/2331abd65af745254c356932b86a1bea485f1f1ac287398e9bf363e0f013e8f1.jpg) +(I) Qualitative results on DressCode +(II) Qualitative results on VITON-HD +Figure 3. Qualitative comparisons of VTON-HandFit with other methods on public datasets: (I) Dresscode and (II) VITON-HD. + +virtual try-on domain as baselines for comparison: LaDIVTON [32], StableVTON [24], OOTDiffusion [48], IDM-VTON [7], and CatVTON [54]. While these methods have made substantial contributions to the virtual try-on field, they exhibit certain limitations in handling hand occlusion issues, which motivated us to propose VTON-HandFit. + +Dataset. Our experiments utilize two publicly virtual try-on datasets: VITON-HD [6] and DressCode [31]. The VITON-HD dataset consists of 13,679 image pairs, divided into 11,647 for training and 2,032 for testing, featuring front-facing upper-body portraits paired with garments at a resolution of $1024 \times 768$ . The DressCode dataset, larger and more varied, contains 48,392 training pairs and 5,400 testing pairs, including full-body portraits and a range of garments (upper-body, lower-body, and dresses) at the same resolution. To enhance our evaluation of the model's capability to address hand occlusion challenges, we build Handfit-3K test with 3,620 images sorted from 26,000 images collected in online retail sites. + +Evaluation metrics. Following previous works [24, 32], we adopt both paired and unpaired evaluation settings. In the paired setting, the quality of the generated images is assessed using Structural Similarity (SSIM) [45], Learned Perceptual Image Patch Similarity (LPIPS) [58], Frechet Inception Distance (FID) [18], and Kernel Inception Distance + +(KID) [3]. For the unpaired setting, we employ FID and KID for comparison. Hand generation quality is assessed using FID-H and KID-H for the hand region. Additionally, we use 2D Mean Per Joint Position Error (MPJPE) [20] to evaluate hand joint position accuracy. + +Implementation details. Our model is developed using the pre-trained StableDiffusion v1.5 [39]. Detailed body dynamics and structural information are captured using DWpose [52] for pose estimation, Detector2 [46] for Dense-sepose segmentation, and HaMeR for detailed hand depth maps and parameters. We use VITON-HD and DressCode datasets for training. Our training strategy is implemented in two phases. In the first phase, we train the Hand-Pose Aggregation Net and Hand-feature Disentanglement Embedding modules while freezing the other components. In the second phase, we finetune the TryonNet and GarmentNet. All models are trained uniformly with a batch size of 32, employing the AdamW [28] optimizer at a learning rate of 5e-5. Training is conducted on 8 NVIDIA V100 GPUs. + +# 4.1. Results on Public Dataset + +Qualitative results. Fig. 3 shows visual comparisons of the VITON-HD and DressCode datasets between VTON-HandFit and other methods. In previous methods, original hand regions are often reused, leading to artifacts such as + +
MethodsDressCodeVITON-HD
PairedUnpairedPairedUnpaired
SSIM ↑LPIPS ↓FID ↓KID ↓FID ↓KID ↓SSIM ↑LPIPS ↓FID ↓KID ↓FID ↓KID ↓
LaDI-VTON (2023)0.90250.07194.86361.55806.84212.33450.87630.09116.60441.06729.40951.6866
StableVTON (2024)------0.86650.08356.85811.25539.58681.4508
OOTDiffusion (2024)0.89750.07253.94970.71986.70191.86300.85130.09646.51860.89619.67331.2061
IDM-VTON (2024)0.92280.04783.80011.20125.61591.55360.88060.07896.33811.32249.61141.6387
IDM-VTON* (2024)0.90070.06244.91702.01766.68672.57930.83160.10587.74912.33819.77121.9323
CatVTON (2024)0.90110.07053.27550.66965.42201.54900.86940.09706.13940.96399.14341.2666
VTON-HandFit (Ours)0.94040.04442.86660.80095.23141.01890.88500.07305.95640.96998.81780.9951
+ +Table 1. Quantitative results on VITON-HD and DressCode datasets. We compare the metrics under both paired (model's clothing is the same as the given cloth image) and unpaired settings (model's clothing differs) with other methods. IDM-VTON* indicates post-processing using HandRefiner. + +![](images/d9fa97f64b899d23b4070c22fbea7fb4d3ec7188994b33d825325e2d3ed392d4.jpg) +Figure 4. Qualitative comparisons of VTON-HandFit with other methods on our Handfit-3K dataset. + +blue outlines around the hands when garment colors significantly differ (third row). In contrast, our model effectively eliminates these residual backgrounds and preserves natural hand poses, enhancing visual fidelity and pose accuracy. + +Quantitative results. Tab. 1 presents the quantitative comparisons of VTON-HandFit against other state-of-the-art methods on VITON-HD and DressCode datasets. By enhancing pose accuracy and the fidelity of hand gestures, our approach achieves markedly improved synthesized image quality. Moreover, VTON-HandFit outperforms in FID and KID scores across both datasets in unpaired settings, underscoring the importance of precise hand pose representation for enhancing image quality in practical applications. + +Limitations of handrefiner. Despite the strides made in hand reconstruction through HandRefiner [29], as a post-processing operation, HandRefiner falls short in its inability to reconstruct the hands without complete shape and fails to depict the same ones with the original model accurately. These deficiencies not only undermine the model's practicality but also compromise the authenticity of its output, thereby substantially diminishing its performance metrics and overall applicability, as shown in Tab. 1. + +![](images/dfcf5b3138264a54321f89f0ea28afbeb99987b7953a88108154f40c5f9ae89e.jpg) +Figure 5. Qualitative comparisons in real-world scenarios. + +![](images/f6b17a074c0e9c0e6efaf9f8797e2ffd923fadad363685b5bcc2457c536c0594.jpg) +Figure 6. Qualitative comparisons of VTON-HandFit with variants on VITON-HD dataset. + +# 4.2. Results on Handfit-3K Dataset + +Qualitative results. Fig. 4 presents a qualitative comparison between VTON-HandFit and other diffusion-based methods. Our method outperforms in preserving detailed garment attributes and achieving natural hand poses. The + +
MethodsPairedUnpaired
SSIM ↑LPIPS ↓FID ↓KID ↓MPJPE ↓FID-H ↓KID-H ↓FID ↓KID ↓MPJPE ↓FID-H ↓KID-H ↓
LaDI-VTON (2023)0.85970.148012.84304.4112107.631523.019917.455816.25067.0065111.154721.137114.1960
StableVTON (2024)0.87720.102211.36754.136054.758622.747016.69914.04465.196657.333422.809916.3022
IDM-VTON (2024)0.88160.09357.71592.128748.637811.54033.972711.29603.463857.344712.77214.4315
OOTDiffusion (2024)0.85540.11848.71042.136383.965512.19843.665313.19694.483986.167813.80114.4479
CatVTON (2024)0.87610.112310.69664.258160.660813.12644.872315.70816.954463.224515.56035.6134
VTON-HandFit (Ours)0.88270.08195.60270.493225.13767.05741.247410.97863.324826.68929.00032.5362
+ +Table 2. Quantitative results on Handfit-3K dataset in the paired and unpaired settings. + +
MethodsSSIM ↑LPIPS ↓FID ↓KID ↓
Baseline0.86190.09236.95491.2651
Baseline+HPA(Densepose)0.86600.08766.89241.1254
Baseline+HPA0.87380.08126.33900.9502
Baseline+HPA+Cstruct0.87840.07876.22251.1186
Baseline+HPA+Cstruct+Cappear0.87870.07926.05151.0084
VTON-HandFit0.88500.07305.95640.9699
+ +Table 3. Comparison of different variants on VITON-HD dataset. + +images generated by VTON-HandFit demonstrate superior consistency in texture, pattern, and fit relative to baseline methods. Furthermore, the hand poses in our images are more consistent with the original. + +Quantitative results. Tab. 2 consolidates the performance of VTON-HandFit against contemporary methods on the Handfit-3K dataset. VTON-HandFit excels in both paired and unpaired settings, showcasing marked advancements in MPJPE, FID-H, and KID-H metrics. This evidences its capability to generate visually compelling images with precise hand pose representation and detailed hand regions. The model's robustness is further affirmed in the unpaired setting, where it consistently leads across all metrics, reinforcing its practical applicability in virtual try-on scenarios with a focus on hand detail fidelity. + +Results of real-world scenarios. To validate the performance of our VTON-HandFit in real-world scenarios, we have collected several cases from natural environments, as illustrated in Fig. 5. It is evident that VTON-HandFit continues to perform robustly in real-world settings, with a noticeable enhancement in hand effects. + +# 4.3. Ablation Study + +Effectiveness of each component. We perform an ablation study to evaluate the contributions of the Hand-Pose Aggregation Net, Hand Structural Embedding $C_{struct}$ , and Hand Appearance Embedding $C_{appear}$ . To facilitate a clear comparison, we introduce several model variants to ascertain the effectiveness of each component. The 'Baseline' model excludes all specialized components. The 'Baseline+HPA (Densepose)' variant incorporates only Densepose for pose control. 'Baseline+HPA' utilizes Densepose, Depth, and DWpose for pose control. 'Baseline+HPA+C_{struct}' excludes the appearance embedding. Lastly, 'Baseline+HPA+C_{struct}+C_{appear}' incorpo + +rates all components. It's important to note that these variants do not include hand-canny constraint loss. Tab. 3 presents a quantitative comparison of our ablation experiments, highlighting the impact of adding different components on the image generation quality. Fig. 6 offers a qualitative analysis, showing the necessity of pose guidance for accurate arm pose replication in the model. By incorporating stronger hand structure and visual guidance, the generation of hands is significantly improved. + +# 5. Conclusion + +We propose VTON-HandFit, a novel model designed for addressing hand occlusion in virtual try-on scenario. Specifically, we have developed a Hand-Pose Aggregation Net by explicitly and adaptively encoding global hand and pose priors using the ControlNet-based structure. To fully exploit the hand-related structure and appearance information, we propose the Hand-feature Disentanglement Embedding module, which explicitly separates the hand priors into hand structure-parametric and visual-appearance features. Lastly, we have customized a hand-canny constraint loss to enhance the learning of structural edge knowledge from the hand template of the model image. Our approach has been extensively evaluated qualitatively and quantitatively, demonstrating its superiority over state-of-the-art virtual try-on models, particularly in cases involving hand pose occlusion. These results pave the way for more complex virtual try-on applications toward real-world scenarios. + +# Acknowledgments + +This work was supported by the National Science Fund for Distinguished Young Scholars (No.62025603) and National Natural Science Foundation of China (No. 62072386), supported by Yunnan Provincial Major S&T Special Plan Project (No. 202402AD080001), Henan Key R&D Project (No. 231111212000), Henan Center for Outstanding Overseas Scientists (No. GZS2022011), Open Foundation of Henan Key Lab of General Aviation Technology (No. ZHKF-230212), Key Lab of Oracle Information Processing of MOE (No. OIP2024E002). We also thank Digital Culture Lab of Tencent SSV for their invaluable contribution during our research. + +# References + +[1] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint arXiv:1607.06450, 2016. 5 +[2] Ankan Kumar Bhunia, Salman Khan, Hisham Cholakkal, Rao Muhammad Anwer, Jorma Laaksonen, Mubarak Shah, and Fahad Shahbaz Khan. Person image synthesis via denoising diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5968-5976, 2023. 3 +[3] Mikołaj Binkowski, Danica J Sutherland, Michael Arbel, and Arthur Gretton. Demystifying mmd gans. arXiv preprint arXiv:1801.01401, 2018. 6 +[4] Shidong Cao, Wenhao Chai, Shengyu Hao, Yanting Zhang, Hangyue Chen, and Gaoang Wang. Diffashion: Reference-based fashion design with structure-aware transfer by diffusion models. IEEE Transactions on Multimedia, 2023. 3 +[5] Mengting Chen, Xi Chen, Zhonghua Zhai, Chen Ju, Xuewen Hong, Jinsong Lan, and Shuai Xiao. Wear-any-way: Manipulable virtual try-on via sparse correspondence alignment. arXiv preprint arXiv:2403.12965, 2024. 3 +[6] Seunghwan Choi, Sunghyun Park, Minsoo Lee, and Jaegul Choo. Viton-hd: High-resolution virtual try-on via misalignment-aware normalization. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14131-14140, 2021. 6 +[7] Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for virtual try-on. ECCV, 2024. 3, 6 +[8] Zheng Chong, Xiao Dong, Haoxiang Li, Shiyue Zhang, Wenqing Zhang, Xujie Zhang, Hanqing Zhao, and Xiaodan Liang. Catvton: Concatenation is all you need for virtual try-on with diffusion models. arXiv preprint arXiv:2407.15886, 2024. 1 +[9] Morelli Davide, Fincato Matteo, Cornia Marcella, Landi Federico, Cesari Fabio, and Cucchiara Rita. Dress code: High-resolution multi-category virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022. 1 +[10] Haoye Dong, Xiaodan Liang, Xiaohui Shen, Bochao Wang, Hanjiang Lai, Jia Zhu, Zhiting Hu, and Jian Yin. Towards multi-pose guided virtual try-on network. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9026-9035, 2019. +[11] Chongjian Ge, Yibing Song, Yuying Ge, Han Yang, Wei Liu, and Ping Luo. Disentangled cycle consistency for highly-realistic virtual try-on. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16928-16937, 2021. 1 +[12] Yuying Ge, Yibing Song, Ruimao Zhang, Chongjian Ge, Wei Liu, and Ping Luo. Parser-free virtual try-on via distilling appearance flows. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8485-8493, 2021. 3 +[13] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and + +Yoshua Bengio. Generative adversarial networks. Communications of the ACM, 63(11):139-144, 2020. 1, 3 +[14] Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang. Taming the power of diffusion models for high-quality virtual try-on with appearance flow. In Proceedings of the 31st ACM International Conference on Multimedia, pages 7599-7607, 2023. 3 +[15] Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S Davis. Viton: An image-based virtual try-on network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7543-7552, 2018. 1 +[16] Xintong Han, Xiaojun Hu, Weilin Huang, and Matthew R Scott. Clothflow: A flow-based model for clothed person generation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 10471-10480, 2019. +[17] Sen He, Yi-Zhe Song, and Tao Xiang. Style-based global appearance flow for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3470-3479, 2022. 1 +[18] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems, 30, 2017. 6 +[19] Xiaobin Hu, Xu Peng, Donghao Luo, Xiaozhong Ji, Jinlong Peng, Zhengkai Jiang, Jiangning Zhang, Taisong Jin, Chengjie Wang, and Rongrong Ji. Diffumatting: Synthesizing arbitrary objects with matting-level annotation. arXiv preprint arXiv:2403.06168, 2024. 4 +[20] Catalin Ionescu, Dragos Papava, Vlad Olaru, and Cristian Sminchisescu. Human3. 6m: Large scale datasets and predictive methods for 3d human sensing in natural environments. IEEE transactions on pattern analysis and machine intelligence, 36(7):1325-1339, 2013. 6 +[21] Thibaut Issenhuth, Jérémie Mary, and Clément Calauzenes. Do not mask what you do not need to mask: a parser-free virtual try-on. In Computer Vision-ECCV 2020: 16th European Conference, Glasgow, UK, August 23-28, 2020, Proceedings, Part XX 16, pages 619-635. Springer, 2020. 3 +[22] Johanna Karras, Aleksander Holynski, Ting-Chun Wang, and Ira Kemelmacher-Shlizerman. Dreampose: Fashion image-to-video synthesis via stable diffusion. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 22623-22633. IEEE, 2023. 3 +[23] Bahjat Kawar, Shiran Zada, Oran Lang, Omer Tov, Huiwen Chang, Tali Dekel, Inbar Mosseri, and Michal Irani. Imagic: Text-based real image editing with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6007-6017, 2023. 4 +[24] Jeongho Kim, Guojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8176-8185, 2024. 3, 6 +[25] Nupur Kumari, Bingliang Zhang, Richard Zhang, Eli Shechtman, and Jun-Yan Zhu. Multi-concept customization of text-to-image diffusion. In Proceedings of the IEEE/CVF + +Conference on Computer Vision and Pattern Recognition, pages 1931-1941, 2023. 4 +[26] Sangyun Lee, Gyojung Gu, Sunghyun Park, Seunghwan Choi, and Jaegul Choo. High-resolution virtual try-on with misalignment and occlusion-handled conditions. In European Conference on Computer Vision, pages 204-219. Springer, 2022. 1, 3 +[27] Yuhan Li, Hao Zhou, Wenxiang Shang, Ran Lin, Xuanhong Chen, and Bingbing Ni. Anyfit: Controllable virtual try-on for any combination of attire across any scenario. arXiv preprint arXiv:2405.18172, 2024. 3 +[28] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 6 +[29] Wenquan Lu, Yufei Xu, Jing Zhang, Chaoyue Wang, and Dacheng Tao. Handrefiner: Refining malformed hands in generated images by diffusion-based conditional inpainting. In ACM Multimedia 2024, 2023. 4, 7 +[30] Yifang Men, Yiming Mao, Yuning Jiang, Wei-Ying Ma, and Zhouhui Lian. Controllable person image synthesis with attribute-decomposed gan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5084-5093, 2020. 3 +[31] Davide Morelli, Matteo Fincato, Marcella Cornia, Federico Landi, Fabio Cesari, and Rita Cucchiara. Dress code: High-resolution multi-category virtual try-on. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2231-2235, 2022. 6 +[32] Davide Morelli, Alberto Baldrati, Giuseppe Cartella, Marcelia Cornia, Marco Bertini, and Rita Cucchiara. Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on. In Proceedings of the 31st ACM International Conference on Multimedia, pages 8580-8589, 2023. 3, 6 +[33] Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4296–4304, 2024. 4 +[34] Maxime Oquab, Timothee Darcet, Theo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 5 +[35] Georgios Pavlakos, Dandan Shan, Ilija Radosavovic, Angjoo Kanazawa, David Fouhey, and Jitendra Malik. Reconstructing hands in 3d with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9826-9836, 2024. 4, 5 +[36] Anton Pelykh, Ozge Mercanoglu Sincan, and Richard Bowden. Giving a hand to diffusion models: a two-stage approach to improving conditional human image generation. arXiv preprint arXiv:2403.10731, 2024. 4 +[37] Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952, 2023. 4 + +[38] Sergey Prokudin, Christoph Lassner, and Javier Romero. Efficient learning on point clouds with basis point sets. In Proceedings of the IEEE/CVF international conference on computer vision, pages 4332-4341, 2019. 5 +[39] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684-10695, 2022. 1, 4, 6 +[40] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22500-22510, 2023. 4 +[41] Chitwan Sahara, William Chan, Huiwen Chang, Chris Lee, Jonathan Ho, Tim Salimans, David Fleet, and Mohammad Norouzi. Palette: Image-to-image diffusion models. In ACM SIGGRAPH 2022 conference proceedings, pages 1-10, 2022. 4 +[42] Dvir Samuel, Rami Ben-Ari, Simon Raviv, Nir Darshan, and Gal Chechik. Generating images of rare concepts using pretrained diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4695-4703, 2024. 4 +[43] Ke Sun, Jian Cao, Qi Wang, Linrui Tian, Xindi Zhang, Lian Zhuo, Bang Zhang, Liefeng Bo, Wenbo Zhou, Weiming Zhang, et al. Outfitanyone: Ultra-high quality virtual try-on for any clothing and any person. arXiv preprint arXiv:2407.16224, 2024. 3 +[44] Chengrui Wang, Pengfei Liu, Min Zhou, Ming Zeng, Xubin Li, Tiezheng Ge, et al. Rhand: Refining malformed hands for generated images with decoupled structure and style guidance. arXiv preprint arXiv:2404.13984, 2024. 4 +[45] Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600-612, 2004. 6 +[46] Yuxin Wu, Alexander Kirillov, Francisco Massa, Wan-Yen Lo, and Ross Girshick. Detectron2. https://github.com/facebookresearch/detectron2, 2019. 6 +[47] Zhenyu Xie, Zaiyu Huang, Xin Dong, Fuwei Zhao, Haoye Dong, Xijin Zhang, Feida Zhu, and Xiaodan Liang. Gqvton: Towards general purpose virtual try-on via collaborative local-flow global-parsing learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 23550-23559, 2023. 3 +[48] Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen. Oot-diffusion: Outfitting fusion based latent diffusion for controllable virtual try-on. arXiv preprint arXiv:2403.01779, 2024.6 +[49] Binxin Yang, Shuyang Gu, Bo Zhang, Ting Zhang, Xuejin Chen, Xiaoyan Sun, Dong Chen, and Fang Wen. Paint by example: Exemplar-based image editing with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18381-18391, 2023. 3 + +[50] Han Yang, Ruimao Zhang, Xiaobao Guo, Wei Liu, Wangmeng Zuo, and Ping Luo. Towards photo-realistic virtual try-on by adaptively generating-preserving image content. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7850-7859, 2020. 1 +[51] Zhijing Yang, Junyang Chen, Yukai Shi, Hao Li, Tianshui Chen, and Liang Lin. Occlumix: Towards de-occlusion virtual try-on by semantically-guided mixup. IEEE Transactions on Multimedia, 25:1477–1488, 2023. 3 +[52] Zhendong Yang, Ailing Zeng, Chun Yuan, and Yu Li. Effective whole-body pose estimation with two-stages distillation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4210-4220, 2023. 6 +[53] Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv preprint arXiv:2308.06721, 2023.4 +[54] Jianhao Zeng, Dan Song, Weizhi Nie, Hongshuo Tian, Tongtong Wang, and An-An Liu. Cat-dm: Controllable accelerated virtual try-on with diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8372-8382, 2024. 6 +[55] Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, Jun Zhu, Lionel M Ni, and Heung-Yeung Shum. Dino: Detr with improved denoising anchor boxes for end-to-end object detection. ICLR, 2023. 2 +[56] Haozhuo Zhang, Bin Zhu, Yu Cao, and Yanbin Hao. Hand1000: Generating realistic hands from text with only 1,000 images. arXiv preprint arXiv:2408.15461, 2024. 4 +[57] Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3836-3847, 2023. 4 +[58] Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 586-595, 2018. 6 +[59] Xujie Zhang, Ente Lin, Xiu Li, Yuxuan Luo, Michael Kampffmeyer, Xin Dong, and Xiaodan Liang. Mmtryon: Multi-modal multi-reference control for high-quality fashion generation. arXiv preprint arXiv:2405.00448, 2024. 3 +[60] Shihao Zhao, Dongdong Chen, Yen-Chun Chen, Jianmin Bao, Shaozhe Hao, Lu Yuan, and Kwan-Yee K Wong. Uni-controlnet: All-in-one control to text-to-image diffusion models. Advances in Neural Information Processing Systems, 36, 2024. 4 +[61] Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Sahara, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two unets. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4606-4615, 2023. 1, 2, 3 \ No newline at end of file diff --git a/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/images.zip b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/images.zip new file mode 100644 index 0000000000000000000000000000000000000000..cdad4e839a47d3162a33cb8f22f35f81ee01412f --- /dev/null +++ b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/images.zip @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8fa5d9afb09c8ad47a5ace8498c4ca6e8b78e45eaa1d97bf9754df9ce08c8855 +size 714006 diff --git a/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/layout.json b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/layout.json new file mode 100644 index 0000000000000000000000000000000000000000..d24a20af72785aa817ffd335f0840a5196fd0196 --- /dev/null +++ b/2025/VTON-HandFit_ Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding/layout.json @@ -0,0 +1,7413 @@ +{ + "pdf_info": [ + { + "para_blocks": [ + { + "bbox": [ + 76, + 102, + 536, + 140 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 76, + 102, + 536, + 140 + ], + "spans": [ + { + "bbox": [ + 76, + 102, + 536, + 140 + ], + "type": "text", + "content": "VTON-HandFit: Virtual Try-on for Arbitrary Hand Pose Guided by Hand Priors Embedding" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "spans": [ + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": "Yujie Liang" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{1,2*}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": ", Xiaobin Hu" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{3*}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": ", Boyuan Jiang" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{3}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": ", Donghao Luo" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{3\\dagger}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": ", Xu Peng" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{3}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": ", Kai Wu" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{3}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": ", Chengming Xu" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{3}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": ", Wenhui Han" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{3}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": ", Taisong Jin" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{1,2\\dagger}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": ", Chengjie Wang" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{3}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": ", Rongrong Ji" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{1,2}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{1}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": "Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, China. \n" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{2}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": "School of Informatics, Xiamen University, China. " + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "inline_equation", + "content": "^{3}" + }, + { + "bbox": [ + 110, + 160, + 502, + 232 + ], + "type": "text", + "content": "Tencent Youtu Lab." + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 121, + 233, + 488, + 274 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 121, + 233, + 488, + 274 + ], + "spans": [ + { + "bbox": [ + 121, + 233, + 488, + 274 + ], + "type": "text", + "content": "liangyujie@stu.xmu.edu.cn,{xiaobinhu,byronjiang,michaelluo,ppxupeng, chengmingxu,jasoncjwang}@tencent.com,wukaiwork,outlook.com, whhan18@gmail.com,{jintaisong,rrji}@xmu.edu.cn" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 151, + 300, + 200, + 313 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 151, + 300, + 200, + 313 + ], + "spans": [ + { + "bbox": [ + 151, + 300, + 200, + 313 + ], + "type": "text", + "content": "Abstract" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 54, + 326, + 297, + 613 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 326, + 297, + 613 + ], + "spans": [ + { + "bbox": [ + 54, + 326, + 297, + 613 + ], + "type": "text", + "content": "Although diffusion-based image virtual try-on has made considerable progress, emerging approaches still struggle to effectively address the issue of hand occlusion (i.e., clothing regions occluded by the hand part), leading to a notable degradation of the try-on performance. To tackle this issue widely existing in real-world scenarios, we propose VTON-HandFit, leveraging the power of hand priors to reconstruct the appearance and structure for hand occlusion cases. Firstly, we tailor a Hand-Pose Aggregation Net using the ControlNet-based structure explicitly and adaptively encoding the global hand and pose priors. Besides, to fully exploit the hand-related structure and appearance information, we propose Hand-feature Disentanglement Embedding module to disentangle the hand priors into the hand structure-parametric and visual-appearance features, and customize a masked cross attention for further decoupled feature embedding. Lastly, we customize a hand-canny constraint loss to better learn the structure edge knowledge from the hand template of model image. VTON-HandFit outperforms the baselines in qualitative and quantitative evaluations on the public dataset and our self-collected hand-occlusion Handfit-3K dataset particularly for the arbitrary hand pose occlusion cases in real-world scenarios. Our project page is at: https://vton-handfit.github.io." + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 55, + 633, + 137, + 645 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 633, + 137, + 645 + ], + "spans": [ + { + "bbox": [ + 55, + 633, + 137, + 645 + ], + "type": "text", + "content": "1. Introduction" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 55, + 653, + 296, + 689 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 653, + 296, + 689 + ], + "spans": [ + { + "bbox": [ + 55, + 653, + 296, + 689 + ], + "type": "text", + "content": "Image-based virtual try-on (VTON) [26] is a widely adopted and highly promising image synthesis technology in the e-commerce industry. Its primary goal is to enhance" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 313, + 301, + 555, + 373 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 301, + 555, + 373 + ], + "spans": [ + { + "bbox": [ + 313, + 301, + 555, + 373 + ], + "type": "text", + "content": "the shopping experience for consumers and minimize advertising expenses for clothing merchants. The VTON task involves generating an image of a human model wearing a specific garment. Over the past few years, countless researchers [9-11, 50] have dedicated significant efforts to achieve more realistic and precise virtual try-on results." + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 376, + 556, + 604 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 376, + 556, + 604 + ], + "spans": [ + { + "bbox": [ + 313, + 376, + 556, + 604 + ], + "type": "text", + "content": "Most recent studies on virtual try-on predominantly utilize generative adversarial networks (GANs) [13] or latent diffusion models (LDMs) [39] for image synthesis. Traditional GAN-based approaches [15-17] often struggle with accurately generating garment folds, realistic lighting and shadows, and lifelike human body representations. Consequently, more recent research has shifted towards LDM-based methods [61], which effectively improve the authenticity of clothed images. However, these researches focus on generating more realistic and natural try-on images while keeping the detailed features of garments. The hands of try-on models are usually strictly on the sides and garments are not occluded by the hands. Such a strict constraint hinders the virtual try-on paving the way toward real-world applications. For instance, as a highly flexible editing task with a wide range of degrees of freedom, digital humans usually meet the occasions that the garments are occluded by the hands, which severely deteriorates the performance of virtual try-on." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 313, + 605, + 557, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 605, + 557, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 605, + 557, + 715 + ], + "type": "text", + "content": "Motivated by the above challenge, basically, there mainly exist two solutions to tackle this arbitrary hand problem: 1). perfect hand parsing model to accurately segment the tiny area among fingers. 2). involving the hand-reconstruction into the try-on processing. As shown in Fig. 1, CatVTON[8] use the parsing model to keep the hand part, leading to unavoidably the remaining artifacts and backgrounds from the model image. The existing parsing models fail to accurately segment the small" + } + ] + } + ], + "index": 11 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "spans": [ + { + "bbox": [ + 65, + 2, + 111, + 34 + ], + "type": "text", + "content": "CVF" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 145, + 0, + 496, + 37 + ], + "type": "header", + "angle": 0, + "lines": [ + { + "bbox": [ + 145, + 0, + 496, + 37 + ], + "spans": [ + { + "bbox": [ + 145, + 0, + 496, + 37 + ], + "type": "text", + "content": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 57, + 693, + 204, + 704 + ], + "type": "page_footnote", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 693, + 204, + 704 + ], + "spans": [ + { + "bbox": [ + 57, + 693, + 204, + 704 + ], + "type": "text", + "content": "* equal contribution, † corresponding author" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 293, + 749, + 318, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 749, + 318, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 749, + 318, + 757 + ], + "type": "text", + "content": "22616" + } + ] + } + ], + "index": 13 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 0 + }, + { + "para_blocks": [ + { + "type": "image", + "bbox": [ + 67, + 70, + 541, + 320 + ], + "blocks": [ + { + "bbox": [ + 67, + 70, + 541, + 320 + ], + "lines": [ + { + "bbox": [ + 67, + 70, + 541, + 320 + ], + "spans": [ + { + "bbox": [ + 67, + 70, + 541, + 320 + ], + "type": "image", + "image_path": "ffb33d5ded6172475685713deedb9529d89316456ef43ff78a83006a4782f928.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 55, + 321, + 555, + 377 + ], + "lines": [ + { + "bbox": [ + 55, + 321, + 555, + 377 + ], + "spans": [ + { + "bbox": [ + 55, + 321, + 555, + 377 + ], + "type": "text", + "content": "Figure 1. Comparison of different models on virtual try-on with hand occlusion. The leftmost three images are reference images, showing the target clothing, the model, and a close-up of the model's hand. Our model excels in preserving hand details and achieving realistic clothing transfer, as highlighted in the red boxes. CatVTON utilizes a parsing model to retain the hand segment, resulting in the inevitable persistence of residual artifacts and backgrounds from the model image. IDM-VTON* indicates post-processing using HandRefiner, which fails to address missing hands and mistakenly alters correctly depicted ones." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "image_caption" + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 388, + 296, + 710 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 388, + 296, + 710 + ], + "spans": [ + { + "bbox": [ + 56, + 388, + 296, + 710 + ], + "type": "text", + "content": "and fragmented finger areas, thus we embrace the latter strategy to involve the hand-reconstruction by disentangling the hand into the appearance and structure knowledge processing in the try-on stage. Then, we propose a novel diffusion-based virtual try-on method, termed as VTON-HandFit, to generate realistic and natural images for arbitrary hand position or hand-occlusion cases with the aid of hand 3D structure-parametric and visual-appearance priors. To embed the pose- and hand-related knowledge, we build a Hand-Pose Net to exploit the body and hand semantic and structure information by getting the body (e.g., DWPose, DensPose) and 3D hand HaMeR-based [61] reconstruction information. Secondly, considering the over-small character of hand that is easily ignored after down-sampling operation, we use the DINO [55] to dig the local appearance and texture knowledge of hand as the hand-visual embedding features. For the structure of hand, we adopt these structure-parametric priors (e.g., 3D MANO-related parameters, handtype and hand-box coordinates) as hand-structure embedding features. After getting structure-parametric and visual-appearance features, we aim to enhance the model's ability to respectively learn hand-related patterns. To achieve this, we inject the hand-structure embedding into the Hand-Pose Net with a primary focus on enhancing the latent features related to hand structure, and the hand-visual embedding into the mainstream U-Net specifically aiming to enhance texture-related features, respec" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 312, + 388, + 555, + 508 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 312, + 388, + 555, + 508 + ], + "spans": [ + { + "bbox": [ + 312, + 388, + 555, + 508 + ], + "type": "text", + "content": "tively. The disentanglement learning of structure and visual appearance enhances the network's ability to efficiently and respectively acquire hand-related knowledge. To prevent the interference of hand-related priors with garment features, we impose a constraint that restricts the hand-related features to operate solely within the hand mask area of the cross-attention mechanism. Finally, we tailor a handcanny constraint loss to more effectively acquire the structural edge knowledge from the hand template of the model image. Our contributions are summarized as follows:" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 310, + 513, + 556, + 712 + ], + "type": "list", + "angle": 0, + "index": 8, + "blocks": [ + { + "bbox": [ + 310, + 513, + 555, + 548 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 310, + 513, + 555, + 548 + ], + "spans": [ + { + "bbox": [ + 310, + 513, + 555, + 548 + ], + "type": "text", + "content": "- We propose VTON-HandFit, a novel virtual try-on architecture considering to solve the Hand pose occlusion problem popularly existing in real-world scenarios." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 310, + 552, + 555, + 612 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 310, + 552, + 555, + 612 + ], + "spans": [ + { + "bbox": [ + 310, + 552, + 555, + 612 + ], + "type": "text", + "content": "- We propose a Hand-Pose Aggregation Net to encode the hand and pose structure priors and also adaptively to adjust the weight of each component. Meanwhile, hand-canny constraint loss is designed to learn the hand structural-edge knowledge of model images." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 310, + 615, + 556, + 698 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 310, + 615, + 556, + 698 + ], + "spans": [ + { + "bbox": [ + 310, + 615, + 556, + 698 + ], + "type": "text", + "content": "- We propose a novel hand-feature disentanglement embedding module disentangling the hand knowledge into the structure-parametric priors (e.g., MANO parameters) and visual-appearance priors (e.g., color). A mask cross-attention mechanism is proposed to constrain the hand-related features only working on the hand mask area to avoid the garment feature disturbance." + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 310, + 701, + 555, + 712 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 310, + 701, + 555, + 712 + ], + "spans": [ + { + "bbox": [ + 310, + 701, + 555, + 712 + ], + "type": "text", + "content": "- We self-collect Handfit-3K test dataset from online retail" + } + ] + } + ], + "index": 7 + } + ], + "sub_type": "text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "text", + "content": "22617" + } + ] + } + ], + "index": 9 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 1 + }, + { + "para_blocks": [ + { + "type": "image", + "bbox": [ + 88, + 71, + 521, + 355 + ], + "blocks": [ + { + "bbox": [ + 88, + 71, + 521, + 355 + ], + "lines": [ + { + "bbox": [ + 88, + 71, + 521, + 355 + ], + "spans": [ + { + "bbox": [ + 88, + 71, + 521, + 355 + ], + "type": "image", + "image_path": "74037f3537d914e6a0e9804d1cc30bb05894de5764f3a14727b1395c8c65a240.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "lines": [ + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "spans": [ + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "text", + "content": "Figure 2. An overview of our VTON-HandFit Network. The network consists of two main components: Hand-feature Disentanglement Embedding and Hand-Pose Aggregation Net. The Hand-feature Disentanglement Embedding module uses the HaMer model to extract hand priors, including hand type " + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "inline_equation", + "content": "T_{h}" + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "text", + "content": ", 3D vertices " + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "inline_equation", + "content": "V_{h}" + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "text", + "content": ", spatial joint locations " + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "inline_equation", + "content": "J_{2d}" + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "text", + "content": ", and joint rotation matrices " + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "inline_equation", + "content": "\\theta_{h}" + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "text", + "content": ". These features are processed by the Hand-Struct processor to derive structural features " + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "inline_equation", + "content": "c_{struc}" + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "text", + "content": ". Simultaneously, hand images cropped using bounding boxes are processed by DINOv2 and the Hand-Appear processor to obtain visual features " + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "inline_equation", + "content": "c_{appear}" + }, + { + "bbox": [ + 54, + 361, + 555, + 437 + ], + "type": "text", + "content": ". The structural and visual features are integrated using mask cross attention. The Hand-Pose Aggregation Net module controls body and hand poses by aggregating DWpose, Densepose, and hand depth maps." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "image_caption" + } + ], + "index": 0 + }, + { + "bbox": [ + 59, + 449, + 296, + 510 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 59, + 449, + 296, + 510 + ], + "spans": [ + { + "bbox": [ + 59, + 449, + 296, + 510 + ], + "type": "text", + "content": "sites to evaluate hand occlusion performance. Extensive qualitative and quantitative evaluations demonstrate our superiority over state-of-the-art virtual try-on models especially for hand pose occlusion cases, paving the road towards more complicated virtual try-on in real-world." + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 524, + 147, + 537 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 524, + 147, + 537 + ], + "spans": [ + { + "bbox": [ + 55, + 524, + 147, + 537 + ], + "type": "text", + "content": "2. Related Works" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 545, + 296, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 545, + 296, + 713 + ], + "spans": [ + { + "bbox": [ + 55, + 545, + 296, + 713 + ], + "type": "text", + "content": "Image-based virtual try-on. In the context of image-based virtual try-on, the objective is to generate a realistic representation of a target person wearing a specific garment, given a pair of corresponding images. A series of studies [26, 30, 47, 51] have been conducted using Generative Adversarial Networks (GANs) [13]. These studies typically involve the initial deformation of the garment to match the body shape of the person, followed by the utilization of a generator to superimpose the deformed garment onto the person image. Despite efforts made to minimize the discrepancy between the warped garment and the person [12, 21, 26], these approaches often lack the ability to generalize well to diverse person images, particularly those with complex backgrounds or intricate poses. There exists" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 313, + 449, + 555, + 497 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 449, + 555, + 497 + ], + "spans": [ + { + "bbox": [ + 313, + 449, + 555, + 497 + ], + "type": "text", + "content": "GAN-based try-on method [26] considers occluded problems and designs the warping operation to avoid the garment warping misalignment, which is totally different from diffusion-based algorithms as an inpainting problem." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 313, + 498, + 556, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 498, + 556, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 498, + 556, + 715 + ], + "type": "text", + "content": "The remarkable generative capabilities of diffusion models have sparked interest in incorporating them into fashion synthesis, particularly in tasks like visual try-on [2, 4, 5, 22, 27, 43, 59]. TryOnDiffusion [61] employs ParallelUNets to preserve garment details and warp the garment for the try-on process. Subsequent studies have treated virtual try-on as an exemplar-based image inpainting problem [49]. These studies involved fine-tuning the inpainting diffusion models using virtual try-on datasets [24, 32] to generate virtual try-on images of superior quality. For instance, such as LADI-VTON [32] and DCI-VTON [14] have emerged, treating clothing as pseudo-words or employing warping networks to integrate garments into pre-trained diffusion models. IDM-VTON [7] incorporates sophisticated attention modules to encode the high-level semantics of the garment and extract low-level features, thereby preserving fine-grained details. While these diffusion methods address issues related to the nature and realism of synthesized images," + } + ] + } + ], + "index": 6 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 749, + 318, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 749, + 318, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 749, + 318, + 757 + ], + "type": "text", + "content": "22618" + } + ] + } + ], + "index": 7 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 2 + }, + { + "para_blocks": [ + { + "bbox": [ + 55, + 72, + 294, + 107 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 72, + 294, + 107 + ], + "spans": [ + { + "bbox": [ + 55, + 72, + 294, + 107 + ], + "type": "text", + "content": "they struggle to meet hard occasions of the hand-based occlusion which very widely exists in the real-world virtual try-on scenarios." + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 109, + 295, + 358 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 109, + 295, + 358 + ], + "spans": [ + { + "bbox": [ + 56, + 109, + 295, + 358 + ], + "type": "text", + "content": "Conditional control of diffusion models. In recent years, latent diffusion models [39] have demonstrated significant success in tasks such as text-to-image generation [19, 25, 37, 40] and image-to-image generation [23, 41]. To achieve more controllable generated results, several studies have introduced conditional control to diffusion models [53, 60]. ControlNet [57] adds spatial conditioning controls (e.g.,anny edges, Hough lines, user scribbles, human key points) to pretrained diffusion models via zero convolutions manner. T2I-Adapter [33] aligns the internal knowledge within T2I models with external control signals, allowing it to be trained based on different conditions. This approach enables the generation of results with rich control and editing effects, particularly in terms of color and structure. In our paper, we focus on the hard occasions of hand-based occlusion in image-based VTON task, and tailor a Hand-Pose Aggregation Net to control the body pose and hand structure. Furthermore, the disengagement knowledge of hand structure and appearance embedding is studied and then these embeddings are respectively injected into the attention mechanism, acting as a mask-based regional control." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 360, + 295, + 622 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 360, + 295, + 622 + ], + "spans": [ + { + "bbox": [ + 56, + 360, + 295, + 622 + ], + "type": "text", + "content": "Reliable human hand generation. Previous studies [42] have acknowledged the problem of misshaped hands that are generated by existing text-to-image diffusion models. Controlling the hand is considerably more challenging compared to controlling the body pose, primarily due to the relative size of the main subject. Besides, diffusion models exhibit significantly improved capability in generating plausible whole-body poses. This difficulty arises from the intricate nature of learning the physical structure and pose of hands from training images, which encompasses extensive deformations and occlusions. Handrefiner [29] and RHanDS [44] are post-processing solutions adopting a hand mesh reconstruction model to rectify malformed hands. Hand-to-Diffusion [36] uses a two-stage approach, first generating hand effects and then producing the entire image. Hand1000 [56] employs a three-stage training process that integrates hand gesture recognition and optimized text embeddings. VTON-HandFit involves the hand corrections into the generation process not as a post-processing and disentangles the hand priors into the structure and appearance feature embedding for virtual try-on arbitrary hand pose in the real-world." + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 633, + 138, + 647 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 633, + 138, + 647 + ], + "spans": [ + { + "bbox": [ + 55, + 633, + 138, + 647 + ], + "type": "text", + "content": "3. Methodology" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 653, + 294, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 653, + 294, + 713 + ], + "spans": [ + { + "bbox": [ + 55, + 653, + 294, + 713 + ], + "type": "text", + "content": "Data collection of Handfit-3K. Considering that the hand of the current main virtual try-on dataset is separate from the body, there rarely exists a great amount of virtual try-on data for arbitrary hand pose and hard hand-occlusion cases. To meet this urgent need in real-world, we aim to" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 313, + 72, + 553, + 216 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 72, + 553, + 216 + ], + "spans": [ + { + "bbox": [ + 313, + 72, + 553, + 216 + ], + "type": "text", + "content": "construct a dataset of complex hand poses that occlude the garment. Specifically, we collect an additional dataset of 26,000 images sourced from online retail sites. These images underwent processing using a parsing model and a hand bounding box via HaMeR model [35] to identify instances of occlusion. We devote significant manual effort to inspecting the quality of the data, removing any erroneous or flawed entries, thereby ensuring the high quality of the virtual try-on dataset. As a result, we obtained a specialized hand occlusion test set comprising 3,620 images, termed as Handfit-3K. This dataset facilitates a targeted evaluation of our method's ability to effectively handle occlusions." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 313, + 223, + 394, + 236 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 223, + 394, + 236 + ], + "spans": [ + { + "bbox": [ + 313, + 223, + 394, + 236 + ], + "type": "text", + "content": "3.1. Preliminary" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 313, + 240, + 553, + 312 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 240, + 553, + 312 + ], + "spans": [ + { + "bbox": [ + 313, + 240, + 553, + 312 + ], + "type": "text", + "content": "Latent Diffusion Models. Our VTON-HandFit is an extension of stable diffusion model, which is a probabilistic model specifically designed to estimate a data distribution " + }, + { + "bbox": [ + 313, + 240, + 553, + 312 + ], + "type": "inline_equation", + "content": "\\mathfrak{p}(\\mathbf{x})" + }, + { + "bbox": [ + 313, + 240, + 553, + 312 + ], + "type": "text", + "content": " by iteratively reducing noise in a normally distributed variable. Given an input image " + }, + { + "bbox": [ + 313, + 240, + 553, + 312 + ], + "type": "inline_equation", + "content": "x_0" + }, + { + "bbox": [ + 313, + 240, + 553, + 312 + ], + "type": "text", + "content": ", the noise estimation process is:" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 319, + 319, + 553, + 335 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 319, + 319, + 553, + 335 + ], + "spans": [ + { + "bbox": [ + 319, + 319, + 553, + 335 + ], + "type": "interline_equation", + "content": "\\mathcal {L} _ {\\text {n o i s e}} = \\mathbb {E} _ {z \\sim \\mathcal {E} (x), C, \\epsilon \\sim \\mathcal {N} (0, 1), t} \\left[ | | \\epsilon - \\epsilon_ {\\theta} (z _ {t}, t, C) | | _ {2} ^ {2} \\right], \\tag {1}", + "image_path": "6f4772e530d95d1307e72e5b98b9e1b06b5841ea4237d0ad06c547ae3d9038ed.jpg" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "spans": [ + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "text", + "content": "where a Variational AutoEncoder " + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "inline_equation", + "content": "\\mathcal{E}" + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "text", + "content": " compress input image " + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "inline_equation", + "content": "x" + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "text", + "content": " into low-dimension latent space " + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "inline_equation", + "content": "z" + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "text", + "content": ". A conditional U-Net denoiser " + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "inline_equation", + "content": "\\epsilon_{\\theta}" + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "text", + "content": " aims to estimate noise " + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "inline_equation", + "content": "\\theta" + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "text", + "content": " in latent space with the aid of timestep " + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "inline_equation", + "content": "t" + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "text", + "content": ", " + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "inline_equation", + "content": "t" + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "text", + "content": "-th noisy latent representation " + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "inline_equation", + "content": "z_{t}" + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "text", + "content": ", and other text-prompt conditions " + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "inline_equation", + "content": "C" + }, + { + "bbox": [ + 313, + 341, + 553, + 401 + ], + "type": "text", + "content": " extracted by text-encoder." + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 407, + 411, + 419 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 407, + 411, + 419 + ], + "spans": [ + { + "bbox": [ + 313, + 407, + 411, + 419 + ], + "type": "text", + "content": "3.2.VTON-HandFit" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "spans": [ + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "text", + "content": "Overview. An overview of our VTON-HandFit is illustrated in Fig. 2. Given a target person image " + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "inline_equation", + "content": "x \\in \\mathbb{R}^{H \\times W \\times 3}" + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "text", + "content": " and an input garment image " + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "inline_equation", + "content": "g \\in \\mathbb{R}^{H \\times W \\times 3}" + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "text", + "content": ", VTON-HandFit aims to generate a realistic outfitted image " + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "inline_equation", + "content": "x_{g} \\in \\mathbb{R}^{H \\times W \\times 3}" + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "text", + "content": ". The garment-agnostic person representation " + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "inline_equation", + "content": "x_{m}" + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "text", + "content": " is masked to erase the garment information by the combination of HumanParsing and HumanPose. Thus, this virtual try-on task is transformed into an example-based inpainting problem to recover the missing mask part. For the input of TryonNet, we simply concatenate the noise image " + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "inline_equation", + "content": "(z_{t})" + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "text", + "content": " and latent garment-agnostic features " + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "inline_equation", + "content": "\\mathcal{E}(x_{m})" + }, + { + "bbox": [ + 313, + 426, + 553, + 713 + ], + "type": "text", + "content": " to avoid the other pose-relevant information disturbances. In Hand-Pose Aggregation Net, we also aggregate the pose-related knowledge including the DWPose, DensePose and 3D reconstruction depth map derived by the HaMeR mesh model. These pose-related priors are detached with TryonNet to explicitly be learned in Hand-Pose Net. DensePose provides more bodysemantic information, DWPose contains abundant skeleton structure, and 3D-hand-derived depth maps demonstrate the pixel-level global-aligned hand position and structure knowledge. These modules complement each other by mutually providing information, and working together to constrain the pose and hand generation. In hand-feature disentanglement embedding, we use the HaMeR model for 3D" + } + ] + } + ], + "index": 11 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 757 + ], + "type": "text", + "content": "22619" + } + ] + } + ], + "index": 12 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 3 + }, + { + "para_blocks": [ + { + "bbox": [ + 55, + 72, + 294, + 179 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 72, + 294, + 179 + ], + "spans": [ + { + "bbox": [ + 55, + 72, + 294, + 179 + ], + "type": "text", + "content": "hand reconstruction and get the structure-parametric priors. These features are processed by the Hand-Struct processor to derive structural features " + }, + { + "bbox": [ + 55, + 72, + 294, + 179 + ], + "type": "inline_equation", + "content": "c_{struc}" + }, + { + "bbox": [ + 55, + 72, + 294, + 179 + ], + "type": "text", + "content": ". Simultaneously, cropped hand images are processed by DINOv2 and the Hand-Appear processor to obtain visual features " + }, + { + "bbox": [ + 55, + 72, + 294, + 179 + ], + "type": "inline_equation", + "content": "c_{appear}" + }, + { + "bbox": [ + 55, + 72, + 294, + 179 + ], + "type": "text", + "content": ". Besides, GarmentNet aims to extract the garment features into the TryonNet mainly by: 1). encoding the garment image into the latent space " + }, + { + "bbox": [ + 55, + 72, + 294, + 179 + ], + "type": "inline_equation", + "content": "\\mathcal{E}_g" + }, + { + "bbox": [ + 55, + 72, + 294, + 179 + ], + "type": "text", + "content": ", 2). using the CLIP to get the image token features for injection." + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 180, + 294, + 288 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 180, + 294, + 288 + ], + "spans": [ + { + "bbox": [ + 55, + 180, + 294, + 288 + ], + "type": "text", + "content": "Hand-Pose Aggregation Net. To aggregate the hand and pose-relevant priors learning, we tailor a Hand-Pose Net to constrain the semantic and skeleton pose, and global pixel-level hand knowledge. Specifically, dwpose, densepose, and render blocks are designed based on the same zero-multilayer structure with different weights to respectively extract the corresponding features. Zero-multilayer structure consists of seven stacking of conv and SiLU activate function with the zero convolution layer at the last layer." + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 102, + 297, + 295, + 310 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 102, + 297, + 295, + 310 + ], + "spans": [ + { + "bbox": [ + 102, + 297, + 295, + 310 + ], + "type": "interline_equation", + "content": "\\mathcal {F} _ {h p} = \\mathcal {F} _ {d w} + \\mathcal {F} _ {d p} + \\mathcal {F} _ {r h} \\times w _ {\\text {h a n d}}, \\tag {2}", + "image_path": "2296c0a9ea0ea77673e0c4df62fa5f648de2783182d78e15d8d3a6f63643fb9e.jpg" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "spans": [ + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "inline_equation", + "content": "\\mathcal{F}_{dw}" + }, + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "text", + "content": ", " + }, + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "inline_equation", + "content": "\\mathcal{F}_{dp}" + }, + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "text", + "content": ", " + }, + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "inline_equation", + "content": "\\mathcal{F}_{rh}" + }, + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "text", + "content": " are the DWPose feature, Densepose feature, and hand feature after the Zero-multilayer structure, and then the summation is implemented on the pixel-level with the adjustable weight " + }, + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "inline_equation", + "content": "w_{\\mathrm{hand}}" + }, + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "text", + "content": " of hand feature " + }, + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "inline_equation", + "content": "\\mathcal{F}_{rh}" + }, + { + "bbox": [ + 55, + 318, + 295, + 413 + ], + "type": "text", + "content": ". To mitigate the impact of detrimental noise on the hidden states of the neural network layers in the trainable Hand-Pose Net during the initial stages of training, we also adopt the zero convolution learning paradigm of ControlNet." + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "spans": [ + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": "Hand structural embedding. The HaMeR [35] model is utilized to reconstruct various MANO parameters and other hand-related information from 2D images, including hand mask " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "M_h \\in \\mathbb{R}^{768 \\times 1024}" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": ", hand types " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "T_h \\in \\mathbb{R}^{n \\times 1}" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": ", hand 3D vertices " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "V_h \\in \\mathbb{R}^{n \\times 778 \\times 3}" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": ", spatial joint locations " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "J_{2d} \\in \\mathbb{R}^{n \\times 21 \\times 2}" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": ", and joint rotation matrices " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "\\theta \\in \\mathbb{R}^{n \\times 16 \\times 3 \\times 3}" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": " where " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "n" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": " represents the number of hands. To accommodate variable numbers of hands as input, we pad the hand parameters to a fixed length " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "n = 2" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": " by adding padding at the end. For hand type " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "T_h" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": ", we use 0 to denote left hands, 1 for right hands, and -1 as a filler for images without hands. Similar padding is applied to other parameters to ensure alignment and correspondence across all dimensions. We apply Basis Point Set (BPS) [38] for dimensionality reduction of hand vertices and encode joint rotations as 6D vectors. And then, we import theses information as well as joint rotations and hand types into the hand-structure processor. The structure of hand-structure processor consists of two parallel blocks: three stacked layers of linear and ReLU activation " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "(L_r)" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": " and two stacked layers of linear and layer normalization " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "(L_n)" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": ". These streamlined hand vertices and joint locations are then fed into " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "L_r" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": ". Concurrently, joint rotations and hand types are transformed via " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "L_n" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": ". The outputs from these processes are aggregated to form the Hand Structural Embedding " + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "inline_equation", + "content": "c_{struc}" + }, + { + "bbox": [ + 55, + 415, + 295, + 714 + ], + "type": "text", + "content": ", encapsulating essential geometric and 3D information for" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 314, + 73, + 430, + 84 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 73, + 430, + 84 + ], + "spans": [ + { + "bbox": [ + 314, + 73, + 430, + 84 + ], + "type": "text", + "content": "precise hand representation." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 337, + 94, + 555, + 107 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 337, + 94, + 555, + 107 + ], + "spans": [ + { + "bbox": [ + 337, + 94, + 555, + 107 + ], + "type": "interline_equation", + "content": "c _ {s t r u c} = \\left[ L _ {r} \\left(\\boldsymbol {V} _ {\\boldsymbol {h}}\\right), L _ {r} \\left(\\boldsymbol {J} _ {\\boldsymbol {2 d}}\\right), L _ {n} \\left(\\boldsymbol {T} _ {\\boldsymbol {h}}\\right), L _ {n} \\left(\\boldsymbol {\\theta} _ {\\boldsymbol {h}}\\right) \\right]. \\tag {3}", + "image_path": "add4ccbfe4f52e02bbb22f9f16fa5985b25de88e57267b8ed4590c51b57ad53b.jpg" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 313, + 117, + 555, + 212 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 117, + 555, + 212 + ], + "spans": [ + { + "bbox": [ + 313, + 117, + 555, + 212 + ], + "type": "text", + "content": "Hand appearance embedding. To improve hand visual detail, we extract features " + }, + { + "bbox": [ + 313, + 117, + 555, + 212 + ], + "type": "inline_equation", + "content": "\\mathcal{F} \\in \\mathbb{R}^{n \\times 1536}" + }, + { + "bbox": [ + 313, + 117, + 555, + 212 + ], + "type": "text", + "content": " from the hand Bounding Box regions using DINOv2 [34]. The Hand-Appear processor (HA) uses a linear layer followed by Layer Normalization [1], culminating in the Hand Appearance Embedding " + }, + { + "bbox": [ + 313, + 117, + 555, + 212 + ], + "type": "inline_equation", + "content": "c_{\\text{appear}}" + }, + { + "bbox": [ + 313, + 117, + 555, + 212 + ], + "type": "text", + "content": ". This process ensures the hand's visual fidelity and detail are meticulously preserved in our representation." + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 394, + 213, + 555, + 225 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 394, + 213, + 555, + 225 + ], + "spans": [ + { + "bbox": [ + 394, + 213, + 555, + 225 + ], + "type": "interline_equation", + "content": "c _ {\\text {a p p e a r}} = \\mathrm {H A} (\\mathcal {F}). \\tag {4}", + "image_path": "bc16cbe8226bf933fc47f232b2ce3c984d2d3122b5775e73b39dabdc3b60fd9d.jpg" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 313, + 231, + 555, + 350 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 231, + 555, + 350 + ], + "spans": [ + { + "bbox": [ + 313, + 231, + 555, + 350 + ], + "type": "text", + "content": "Mask-based regional control. Our method introduces a mask-based regional control to enhance hand detail generation and maintain quality in virtual try-on. Utilizing the hand mask " + }, + { + "bbox": [ + 313, + 231, + 555, + 350 + ], + "type": "inline_equation", + "content": "M_h" + }, + { + "bbox": [ + 313, + 231, + 555, + 350 + ], + "type": "text", + "content": ", we modulate Cross-Attention queries, seamlessly integrating Hand Structural Embedding within Hand-Pose Net's Cross-Attention module. Concurrently, Hand Appearance Embedding is adeptly fused into the Try-onNet architecture through mask cross-attention, ensuring precise detail preservation. The mask cross-attention can be represented as:" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 368, + 361, + 554, + 403 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 368, + 361, + 554, + 403 + ], + "spans": [ + { + "bbox": [ + 368, + 361, + 554, + 403 + ], + "type": "interline_equation", + "content": "\\begin{array}{l} Z = \\operatorname {A t t e n t i o n} (Q, K, V, M _ {h}) \\\\ = \\operatorname {S o f t m a x} \\left(\\frac {\\left(Q K ^ {\\top}\\right) \\odot M _ {h}}{\\sqrt {d}}\\right) V, \\tag {5} \\\\ \\end{array}", + "image_path": "84cc5467a6306869c004c1b765ebe887e325befde0403fcc1e6c4071745e45cd.jpg" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 313, + 411, + 554, + 446 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 411, + 554, + 446 + ], + "spans": [ + { + "bbox": [ + 313, + 411, + 554, + 446 + ], + "type": "text", + "content": "where " + }, + { + "bbox": [ + 313, + 411, + 554, + 446 + ], + "type": "inline_equation", + "content": "Q, K" + }, + { + "bbox": [ + 313, + 411, + 554, + 446 + ], + "type": "text", + "content": " and " + }, + { + "bbox": [ + 313, + 411, + 554, + 446 + ], + "type": "inline_equation", + "content": "V" + }, + { + "bbox": [ + 313, + 411, + 554, + 446 + ], + "type": "text", + "content": " are the query, key, and value matrices of the attention operation for hand appearance embedding cross-attention, " + }, + { + "bbox": [ + 313, + 411, + 554, + 446 + ], + "type": "inline_equation", + "content": "M_h" + }, + { + "bbox": [ + 313, + 411, + 554, + 446 + ], + "type": "text", + "content": " represents the hand mask." + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "spans": [ + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "text", + "content": "Hand canny shape constraint. During the training process, " + }, + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "inline_equation", + "content": "\\mathcal{L}_{hand}" + }, + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "text", + "content": " is implemented to refine the clarity and precision of hand edges. Specifically, for a small " + }, + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "inline_equation", + "content": "R_{t} < T" + }, + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "text", + "content": ", we employ one-step denoising process prior to sampling " + }, + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "inline_equation", + "content": "x_0" + }, + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "text", + "content": " for the direct prediction of " + }, + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "inline_equation", + "content": "z_0" + }, + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "text", + "content": " from " + }, + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "inline_equation", + "content": "z_{t}" + }, + { + "bbox": [ + 313, + 447, + 554, + 518 + ], + "type": "text", + "content": " and the model-predicted noise, using Variational Autoencoder for decoder." + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 365, + 528, + 555, + 554 + ], + "type": "interline_equation", + "angle": 0, + "lines": [ + { + "bbox": [ + 365, + 528, + 555, + 554 + ], + "spans": [ + { + "bbox": [ + 365, + 528, + 555, + 554 + ], + "type": "interline_equation", + "content": "z _ {0} = \\frac {z _ {t} - \\sqrt {1 - \\bar {\\alpha} _ {t}} \\epsilon_ {\\theta}}{\\sqrt {\\bar {\\alpha} _ {t}}}, \\quad t \\leq R _ {t}. \\tag {6}", + "image_path": "c0bebaa79890b9732c14bf0b28384842ce42d6212b306ce66d7cfc2e19bec6d4.jpg" + } + ] + } + ], + "index": 13 + }, + { + "bbox": [ + 313, + 562, + 554, + 658 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 562, + 554, + 658 + ], + "spans": [ + { + "bbox": [ + 313, + 562, + 554, + 658 + ], + "type": "text", + "content": "Hand regions are extracted from the generated images using bounding box cropping and are processed with the Canny edge detection operator. These regions are then compared to their corresponding areas in the Ground Truth images using Mean Squared Error (MSE) loss. This loss function is specifically designed to sharpen the model's focus on improving edge details and clarity in the hand regions of the generated images." + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 314, + 669, + 391, + 682 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 314, + 669, + 391, + 682 + ], + "spans": [ + { + "bbox": [ + 314, + 669, + 391, + 682 + ], + "type": "text", + "content": "4. Experiment" + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 313, + 689, + 554, + 712 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 689, + 554, + 712 + ], + "spans": [ + { + "bbox": [ + 313, + 689, + 554, + 712 + ], + "type": "text", + "content": "Baselines. To comprehensively evaluate our approach, we selected five state-of-the-art diffusion-based methods in the" + } + ] + } + ], + "index": 16 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "text", + "content": "22620" + } + ] + } + ], + "index": 17 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 4 + }, + { + "para_blocks": [ + { + "type": "image", + "bbox": [ + 118, + 72, + 492, + 223 + ], + "blocks": [ + { + "bbox": [ + 118, + 72, + 492, + 223 + ], + "lines": [ + { + "bbox": [ + 118, + 72, + 492, + 223 + ], + "spans": [ + { + "bbox": [ + 118, + 72, + 492, + 223 + ], + "type": "image", + "image_path": "87c2702bedbfbc89cfbca50f4f4eb62cfa75b3a211cc8e5d581619b6cc4c3ee6.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "image_body" + } + ], + "index": 0 + }, + { + "type": "image", + "bbox": [ + 118, + 232, + 492, + 373 + ], + "blocks": [ + { + "bbox": [ + 253, + 224, + 362, + 232 + ], + "lines": [ + { + "bbox": [ + 253, + 224, + 362, + 232 + ], + "spans": [ + { + "bbox": [ + 253, + 224, + 362, + 232 + ], + "type": "text", + "content": "(I) Qualitative results on DressCode" + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 118, + 232, + 492, + 373 + ], + "lines": [ + { + "bbox": [ + 118, + 232, + 492, + 373 + ], + "spans": [ + { + "bbox": [ + 118, + 232, + 492, + 373 + ], + "type": "image", + "image_path": "2331abd65af745254c356932b86a1bea485f1f1ac287398e9bf363e0f013e8f1.jpg" + } + ] + } + ], + "index": 2, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 249, + 374, + 362, + 384 + ], + "lines": [ + { + "bbox": [ + 249, + 374, + 362, + 384 + ], + "spans": [ + { + "bbox": [ + 249, + 374, + 362, + 384 + ], + "type": "text", + "content": "(II) Qualitative results on VITON-HD" + } + ] + } + ], + "index": 3, + "angle": 0, + "type": "image_caption" + }, + { + "bbox": [ + 72, + 387, + 536, + 399 + ], + "lines": [ + { + "bbox": [ + 72, + 387, + 536, + 399 + ], + "spans": [ + { + "bbox": [ + 72, + 387, + 536, + 399 + ], + "type": "text", + "content": "Figure 3. Qualitative comparisons of VTON-HandFit with other methods on public datasets: (I) Dresscode and (II) VITON-HD." + } + ] + } + ], + "index": 4, + "angle": 0, + "type": "image_caption" + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 408, + 295, + 480 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 408, + 295, + 480 + ], + "spans": [ + { + "bbox": [ + 55, + 408, + 295, + 480 + ], + "type": "text", + "content": "virtual try-on domain as baselines for comparison: LaDIVTON [32], StableVTON [24], OOTDiffusion [48], IDM-VTON [7], and CatVTON [54]. While these methods have made substantial contributions to the virtual try-on field, they exhibit certain limitations in handling hand occlusion issues, which motivated us to propose VTON-HandFit." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 483, + 295, + 639 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 483, + 295, + 639 + ], + "spans": [ + { + "bbox": [ + 55, + 483, + 295, + 639 + ], + "type": "text", + "content": "Dataset. Our experiments utilize two publicly virtual try-on datasets: VITON-HD [6] and DressCode [31]. The VITON-HD dataset consists of 13,679 image pairs, divided into 11,647 for training and 2,032 for testing, featuring front-facing upper-body portraits paired with garments at a resolution of " + }, + { + "bbox": [ + 55, + 483, + 295, + 639 + ], + "type": "inline_equation", + "content": "1024 \\times 768" + }, + { + "bbox": [ + 55, + 483, + 295, + 639 + ], + "type": "text", + "content": ". The DressCode dataset, larger and more varied, contains 48,392 training pairs and 5,400 testing pairs, including full-body portraits and a range of garments (upper-body, lower-body, and dresses) at the same resolution. To enhance our evaluation of the model's capability to address hand occlusion challenges, we build Handfit-3K test with 3,620 images sorted from 26,000 images collected in online retail sites." + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 55, + 642, + 295, + 714 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 642, + 295, + 714 + ], + "spans": [ + { + "bbox": [ + 55, + 642, + 295, + 714 + ], + "type": "text", + "content": "Evaluation metrics. Following previous works [24, 32], we adopt both paired and unpaired evaluation settings. In the paired setting, the quality of the generated images is assessed using Structural Similarity (SSIM) [45], Learned Perceptual Image Patch Similarity (LPIPS) [58], Frechet Inception Distance (FID) [18], and Kernel Inception Distance" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 313, + 408, + 555, + 468 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 408, + 555, + 468 + ], + "spans": [ + { + "bbox": [ + 313, + 408, + 555, + 468 + ], + "type": "text", + "content": "(KID) [3]. For the unpaired setting, we employ FID and KID for comparison. Hand generation quality is assessed using FID-H and KID-H for the hand region. Additionally, we use 2D Mean Per Joint Position Error (MPJPE) [20] to evaluate hand joint position accuracy." + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 313, + 469, + 556, + 636 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 469, + 556, + 636 + ], + "spans": [ + { + "bbox": [ + 313, + 469, + 556, + 636 + ], + "type": "text", + "content": "Implementation details. Our model is developed using the pre-trained StableDiffusion v1.5 [39]. Detailed body dynamics and structural information are captured using DWpose [52] for pose estimation, Detector2 [46] for Dense-sepose segmentation, and HaMeR for detailed hand depth maps and parameters. We use VITON-HD and DressCode datasets for training. Our training strategy is implemented in two phases. In the first phase, we train the Hand-Pose Aggregation Net and Hand-feature Disentanglement Embedding modules while freezing the other components. In the second phase, we finetune the TryonNet and GarmentNet. All models are trained uniformly with a batch size of 32, employing the AdamW [28] optimizer at a learning rate of 5e-5. Training is conducted on 8 NVIDIA V100 GPUs." + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 647, + 456, + 658 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 647, + 456, + 658 + ], + "spans": [ + { + "bbox": [ + 313, + 647, + 456, + 658 + ], + "type": "text", + "content": "4.1. Results on Public Dataset" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 313, + 665, + 555, + 713 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 665, + 555, + 713 + ], + "spans": [ + { + "bbox": [ + 313, + 665, + 555, + 713 + ], + "type": "text", + "content": "Qualitative results. Fig. 3 shows visual comparisons of the VITON-HD and DressCode datasets between VTON-HandFit and other methods. In previous methods, original hand regions are often reused, leading to artifacts such as" + } + ] + } + ], + "index": 11 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "text", + "content": "22621" + } + ] + } + ], + "index": 12 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 5 + }, + { + "para_blocks": [ + { + "type": "table", + "bbox": [ + 56, + 70, + 553, + 193 + ], + "blocks": [ + { + "bbox": [ + 56, + 70, + 553, + 193 + ], + "lines": [ + { + "bbox": [ + 56, + 70, + 553, + 193 + ], + "spans": [ + { + "bbox": [ + 56, + 70, + 553, + 193 + ], + "type": "table", + "html": "
MethodsDressCodeVITON-HD
PairedUnpairedPairedUnpaired
SSIM ↑LPIPS ↓FID ↓KID ↓FID ↓KID ↓SSIM ↑LPIPS ↓FID ↓KID ↓FID ↓KID ↓
LaDI-VTON (2023)0.90250.07194.86361.55806.84212.33450.87630.09116.60441.06729.40951.6866
StableVTON (2024)------0.86650.08356.85811.25539.58681.4508
OOTDiffusion (2024)0.89750.07253.94970.71986.70191.86300.85130.09646.51860.89619.67331.2061
IDM-VTON (2024)0.92280.04783.80011.20125.61591.55360.88060.07896.33811.32249.61141.6387
IDM-VTON* (2024)0.90070.06244.91702.01766.68672.57930.83160.10587.74912.33819.77121.9323
CatVTON (2024)0.90110.07053.27550.66965.42201.54900.86940.09706.13940.96399.14341.2666
VTON-HandFit (Ours)0.94040.04442.86660.80095.23141.01890.88500.07305.95640.96998.81780.9951
", + "image_path": "a0ee4e0f278accf5d87660f54d1f89d64502e161d0a37911326d8051d4ea476d.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "table_body" + } + ], + "index": 0 + }, + { + "bbox": [ + 55, + 196, + 555, + 229 + ], + "lines": [ + { + "bbox": [ + 55, + 196, + 555, + 229 + ], + "spans": [ + { + "bbox": [ + 55, + 196, + 555, + 229 + ], + "type": "text", + "content": "Table 1. Quantitative results on VITON-HD and DressCode datasets. We compare the metrics under both paired (model's clothing is the same as the given cloth image) and unpaired settings (model's clothing differs) with other methods. IDM-VTON* indicates post-processing using HandRefiner." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "text" + }, + { + "type": "image", + "bbox": [ + 92, + 236, + 517, + 407 + ], + "blocks": [ + { + "bbox": [ + 92, + 236, + 517, + 407 + ], + "lines": [ + { + "bbox": [ + 92, + 236, + 517, + 407 + ], + "spans": [ + { + "bbox": [ + 92, + 236, + 517, + 407 + ], + "type": "image", + "image_path": "d9fa97f64b899d23b4070c22fbea7fb4d3ec7188994b33d825325e2d3ed392d4.jpg" + } + ] + } + ], + "index": 2, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 122, + 411, + 487, + 423 + ], + "lines": [ + { + "bbox": [ + 122, + 411, + 487, + 423 + ], + "spans": [ + { + "bbox": [ + 122, + 411, + 487, + 423 + ], + "type": "text", + "content": "Figure 4. Qualitative comparisons of VTON-HandFit with other methods on our Handfit-3K dataset." + } + ] + } + ], + "index": 3, + "angle": 0, + "type": "image_caption" + } + ], + "index": 2 + }, + { + "bbox": [ + 55, + 432, + 296, + 482 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 432, + 296, + 482 + ], + "spans": [ + { + "bbox": [ + 55, + 432, + 296, + 482 + ], + "type": "text", + "content": "blue outlines around the hands when garment colors significantly differ (third row). In contrast, our model effectively eliminates these residual backgrounds and preserves natural hand poses, enhancing visual fidelity and pose accuracy." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 55, + 489, + 296, + 598 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 489, + 296, + 598 + ], + "spans": [ + { + "bbox": [ + 55, + 489, + 296, + 598 + ], + "type": "text", + "content": "Quantitative results. Tab. 1 presents the quantitative comparisons of VTON-HandFit against other state-of-the-art methods on VITON-HD and DressCode datasets. By enhancing pose accuracy and the fidelity of hand gestures, our approach achieves markedly improved synthesized image quality. Moreover, VTON-HandFit outperforms in FID and KID scores across both datasets in unpaired settings, underscoring the importance of precise hand pose representation for enhancing image quality in practical applications." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 605, + 296, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 605, + 296, + 715 + ], + "spans": [ + { + "bbox": [ + 55, + 605, + 296, + 715 + ], + "type": "text", + "content": "Limitations of handrefiner. Despite the strides made in hand reconstruction through HandRefiner [29], as a post-processing operation, HandRefiner falls short in its inability to reconstruct the hands without complete shape and fails to depict the same ones with the original model accurately. These deficiencies not only undermine the model's practicality but also compromise the authenticity of its output, thereby substantially diminishing its performance metrics and overall applicability, as shown in Tab. 1." + } + ] + } + ], + "index": 6 + }, + { + "type": "image", + "bbox": [ + 324, + 433, + 545, + 540 + ], + "blocks": [ + { + "bbox": [ + 324, + 433, + 545, + 540 + ], + "lines": [ + { + "bbox": [ + 324, + 433, + 545, + 540 + ], + "spans": [ + { + "bbox": [ + 324, + 433, + 545, + 540 + ], + "type": "image", + "image_path": "dfcf5b3138264a54321f89f0ea28afbeb99987b7953a88108154f40c5f9ae89e.jpg" + } + ] + } + ], + "index": 7, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 328, + 544, + 539, + 555 + ], + "lines": [ + { + "bbox": [ + 328, + 544, + 539, + 555 + ], + "spans": [ + { + "bbox": [ + 328, + 544, + 539, + 555 + ], + "type": "text", + "content": "Figure 5. Qualitative comparisons in real-world scenarios." + } + ] + } + ], + "index": 8, + "angle": 0, + "type": "image_caption" + } + ], + "index": 7 + }, + { + "type": "image", + "bbox": [ + 323, + 559, + 541, + 609 + ], + "blocks": [ + { + "bbox": [ + 323, + 559, + 541, + 609 + ], + "lines": [ + { + "bbox": [ + 323, + 559, + 541, + 609 + ], + "spans": [ + { + "bbox": [ + 323, + 559, + 541, + 609 + ], + "type": "image", + "image_path": "f6b17a074c0e9c0e6efaf9f8797e2ffd923fadad363685b5bcc2457c536c0594.jpg" + } + ] + } + ], + "index": 9, + "angle": 0, + "type": "image_body" + }, + { + "bbox": [ + 313, + 614, + 555, + 636 + ], + "lines": [ + { + "bbox": [ + 313, + 614, + 555, + 636 + ], + "spans": [ + { + "bbox": [ + 313, + 614, + 555, + 636 + ], + "type": "text", + "content": "Figure 6. Qualitative comparisons of VTON-HandFit with variants on VITON-HD dataset." + } + ] + } + ], + "index": 10, + "angle": 0, + "type": "image_caption" + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 647, + 480, + 658 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 647, + 480, + 658 + ], + "spans": [ + { + "bbox": [ + 313, + 647, + 480, + 658 + ], + "type": "text", + "content": "4.2. Results on Handfit-3K Dataset" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 313, + 665, + 555, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 665, + 555, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 665, + 555, + 715 + ], + "type": "text", + "content": "Qualitative results. Fig. 4 presents a qualitative comparison between VTON-HandFit and other diffusion-based methods. Our method outperforms in preserving detailed garment attributes and achieving natural hand poses. The" + } + ] + } + ], + "index": 12 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 318, + 758 + ], + "type": "text", + "content": "22622" + } + ] + } + ], + "index": 13 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 6 + }, + { + "para_blocks": [ + { + "type": "table", + "bbox": [ + 56, + 70, + 553, + 162 + ], + "blocks": [ + { + "bbox": [ + 56, + 70, + 553, + 162 + ], + "lines": [ + { + "bbox": [ + 56, + 70, + 553, + 162 + ], + "spans": [ + { + "bbox": [ + 56, + 70, + 553, + 162 + ], + "type": "table", + "html": "
MethodsPairedUnpaired
SSIM ↑LPIPS ↓FID ↓KID ↓MPJPE ↓FID-H ↓KID-H ↓FID ↓KID ↓MPJPE ↓FID-H ↓KID-H ↓
LaDI-VTON (2023)0.85970.148012.84304.4112107.631523.019917.455816.25067.0065111.154721.137114.1960
StableVTON (2024)0.87720.102211.36754.136054.758622.747016.69914.04465.196657.333422.809916.3022
IDM-VTON (2024)0.88160.09357.71592.128748.637811.54033.972711.29603.463857.344712.77214.4315
OOTDiffusion (2024)0.85540.11848.71042.136383.965512.19843.665313.19694.483986.167813.80114.4479
CatVTON (2024)0.87610.112310.69664.258160.660813.12644.872315.70816.954463.224515.56035.6134
VTON-HandFit (Ours)0.88270.08195.60270.493225.13767.05741.247410.97863.324826.68929.00032.5362
", + "image_path": "6d1d9da32f0888bb5f3c5577af8167412fc47605c96aa3c73fa9eab36246fffc.jpg" + } + ] + } + ], + "index": 0, + "angle": 0, + "type": "table_body" + } + ], + "index": 0 + }, + { + "type": "table", + "bbox": [ + 56, + 190, + 294, + 262 + ], + "blocks": [ + { + "bbox": [ + 147, + 165, + 462, + 177 + ], + "lines": [ + { + "bbox": [ + 147, + 165, + 462, + 177 + ], + "spans": [ + { + "bbox": [ + 147, + 165, + 462, + 177 + ], + "type": "text", + "content": "Table 2. Quantitative results on Handfit-3K dataset in the paired and unpaired settings." + } + ] + } + ], + "index": 1, + "angle": 0, + "type": "table_caption" + }, + { + "bbox": [ + 56, + 190, + 294, + 262 + ], + "lines": [ + { + "bbox": [ + 56, + 190, + 294, + 262 + ], + "spans": [ + { + "bbox": [ + 56, + 190, + 294, + 262 + ], + "type": "table", + "html": "
MethodsSSIM ↑LPIPS ↓FID ↓KID ↓
Baseline0.86190.09236.95491.2651
Baseline+HPA(Densepose)0.86600.08766.89241.1254
Baseline+HPA0.87380.08126.33900.9502
Baseline+HPA+Cstruct0.87840.07876.22251.1186
Baseline+HPA+Cstruct+Cappear0.87870.07926.05151.0084
VTON-HandFit0.88500.07305.95640.9699
", + "image_path": "7ed934ed683f776075d78c6cdfe0ddabd4270fa653ef7d30607c1c6b3db7f1f3.jpg" + } + ] + } + ], + "index": 2, + "angle": 0, + "type": "table_body" + } + ], + "index": 2 + }, + { + "bbox": [ + 56, + 265, + 294, + 277 + ], + "lines": [ + { + "bbox": [ + 56, + 265, + 294, + 277 + ], + "spans": [ + { + "bbox": [ + 56, + 265, + 294, + 277 + ], + "type": "text", + "content": "Table 3. Comparison of different variants on VITON-HD dataset." + } + ] + } + ], + "index": 3, + "angle": 0, + "type": "text" + }, + { + "bbox": [ + 55, + 292, + 295, + 339 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 292, + 295, + 339 + ], + "spans": [ + { + "bbox": [ + 55, + 292, + 295, + 339 + ], + "type": "text", + "content": "images generated by VTON-HandFit demonstrate superior consistency in texture, pattern, and fit relative to baseline methods. Furthermore, the hand poses in our images are more consistent with the original." + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 54, + 340, + 295, + 471 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 54, + 340, + 295, + 471 + ], + "spans": [ + { + "bbox": [ + 54, + 340, + 295, + 471 + ], + "type": "text", + "content": "Quantitative results. Tab. 2 consolidates the performance of VTON-HandFit against contemporary methods on the Handfit-3K dataset. VTON-HandFit excels in both paired and unpaired settings, showcasing marked advancements in MPJPE, FID-H, and KID-H metrics. This evidences its capability to generate visually compelling images with precise hand pose representation and detailed hand regions. The model's robustness is further affirmed in the unpaired setting, where it consistently leads across all metrics, reinforcing its practical applicability in virtual try-on scenarios with a focus on hand detail fidelity." + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 55, + 472, + 295, + 544 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 472, + 295, + 544 + ], + "spans": [ + { + "bbox": [ + 55, + 472, + 295, + 544 + ], + "type": "text", + "content": "Results of real-world scenarios. To validate the performance of our VTON-HandFit in real-world scenarios, we have collected several cases from natural environments, as illustrated in Fig. 5. It is evident that VTON-HandFit continues to perform robustly in real-world settings, with a noticeable enhancement in hand effects." + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 55, + 552, + 149, + 564 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 552, + 149, + 564 + ], + "spans": [ + { + "bbox": [ + 55, + 552, + 149, + 564 + ], + "type": "text", + "content": "4.3. Ablation Study" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 55, + 570, + 295, + 714 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 55, + 570, + 295, + 714 + ], + "spans": [ + { + "bbox": [ + 55, + 570, + 295, + 714 + ], + "type": "text", + "content": "Effectiveness of each component. We perform an ablation study to evaluate the contributions of the Hand-Pose Aggregation Net, Hand Structural Embedding " + }, + { + "bbox": [ + 55, + 570, + 295, + 714 + ], + "type": "inline_equation", + "content": "C_{struct}" + }, + { + "bbox": [ + 55, + 570, + 295, + 714 + ], + "type": "text", + "content": ", and Hand Appearance Embedding " + }, + { + "bbox": [ + 55, + 570, + 295, + 714 + ], + "type": "inline_equation", + "content": "C_{appear}" + }, + { + "bbox": [ + 55, + 570, + 295, + 714 + ], + "type": "text", + "content": ". To facilitate a clear comparison, we introduce several model variants to ascertain the effectiveness of each component. The 'Baseline' model excludes all specialized components. The 'Baseline+HPA (Densepose)' variant incorporates only Densepose for pose control. 'Baseline+HPA' utilizes Densepose, Depth, and DWpose for pose control. 'Baseline+HPA+C_{struct}' excludes the appearance embedding. Lastly, 'Baseline+HPA+C_{struct}+C_{appear}' incorpo" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 313, + 192, + 555, + 300 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 192, + 555, + 300 + ], + "spans": [ + { + "bbox": [ + 313, + 192, + 555, + 300 + ], + "type": "text", + "content": "rates all components. It's important to note that these variants do not include hand-canny constraint loss. Tab. 3 presents a quantitative comparison of our ablation experiments, highlighting the impact of adding different components on the image generation quality. Fig. 6 offers a qualitative analysis, showing the necessity of pose guidance for accurate arm pose replication in the model. By incorporating stronger hand structure and visual guidance, the generation of hands is significantly improved." + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 313, + 312, + 388, + 324 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 312, + 388, + 324 + ], + "spans": [ + { + "bbox": [ + 313, + 312, + 388, + 324 + ], + "type": "text", + "content": "5. Conclusion" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 313, + 332, + 555, + 537 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 332, + 555, + 537 + ], + "spans": [ + { + "bbox": [ + 313, + 332, + 555, + 537 + ], + "type": "text", + "content": "We propose VTON-HandFit, a novel model designed for addressing hand occlusion in virtual try-on scenario. Specifically, we have developed a Hand-Pose Aggregation Net by explicitly and adaptively encoding global hand and pose priors using the ControlNet-based structure. To fully exploit the hand-related structure and appearance information, we propose the Hand-feature Disentanglement Embedding module, which explicitly separates the hand priors into hand structure-parametric and visual-appearance features. Lastly, we have customized a hand-canny constraint loss to enhance the learning of structural edge knowledge from the hand template of the model image. Our approach has been extensively evaluated qualitatively and quantitatively, demonstrating its superiority over state-of-the-art virtual try-on models, particularly in cases involving hand pose occlusion. These results pave the way for more complex virtual try-on applications toward real-world scenarios." + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 313, + 549, + 411, + 562 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 549, + 411, + 562 + ], + "spans": [ + { + "bbox": [ + 313, + 549, + 411, + 562 + ], + "type": "text", + "content": "Acknowledgments" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 313, + 570, + 556, + 715 + ], + "type": "text", + "angle": 0, + "lines": [ + { + "bbox": [ + 313, + 570, + 556, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 570, + 556, + 715 + ], + "type": "text", + "content": "This work was supported by the National Science Fund for Distinguished Young Scholars (No.62025603) and National Natural Science Foundation of China (No. 62072386), supported by Yunnan Provincial Major S&T Special Plan Project (No. 202402AD080001), Henan Key R&D Project (No. 231111212000), Henan Center for Outstanding Overseas Scientists (No. GZS2022011), Open Foundation of Henan Key Lab of General Aviation Technology (No. ZHKF-230212), Key Lab of Oracle Information Processing of MOE (No. OIP2024E002). We also thank Digital Culture Lab of Tencent SSV for their invaluable contribution during our research." + } + ] + } + ], + "index": 13 + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 748, + 317, + 758 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 748, + 317, + 758 + ], + "spans": [ + { + "bbox": [ + 293, + 748, + 317, + 758 + ], + "type": "text", + "content": "22623" + } + ] + } + ], + "index": 14 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 7 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 71, + 115, + 83 + ], + "type": "title", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 71, + 115, + 83 + ], + "spans": [ + { + "bbox": [ + 56, + 71, + 115, + 83 + ], + "type": "text", + "content": "References" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 57, + 91, + 296, + 713 + ], + "type": "list", + "angle": 0, + "index": 14, + "blocks": [ + { + "bbox": [ + 61, + 91, + 296, + 123 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 61, + 91, + 296, + 123 + ], + "spans": [ + { + "bbox": [ + 61, + 91, + 296, + 123 + ], + "type": "text", + "content": "[1] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint arXiv:1607.06450, 2016. 5" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 61, + 125, + 296, + 191 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 61, + 125, + 296, + 191 + ], + "spans": [ + { + "bbox": [ + 61, + 125, + 296, + 191 + ], + "type": "text", + "content": "[2] Ankan Kumar Bhunia, Salman Khan, Hisham Cholakkal, Rao Muhammad Anwer, Jorma Laaksonen, Mubarak Shah, and Fahad Shahbaz Khan. Person image synthesis via denoising diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5968-5976, 2023. 3" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 62, + 194, + 295, + 225 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 194, + 295, + 225 + ], + "spans": [ + { + "bbox": [ + 62, + 194, + 295, + 225 + ], + "type": "text", + "content": "[3] Mikołaj Binkowski, Danica J Sutherland, Michael Arbel, and Arthur Gretton. Demystifying mmd gans. arXiv preprint arXiv:1801.01401, 2018. 6" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 62, + 228, + 295, + 271 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 228, + 295, + 271 + ], + "spans": [ + { + "bbox": [ + 62, + 228, + 295, + 271 + ], + "type": "text", + "content": "[4] Shidong Cao, Wenhao Chai, Shengyu Hao, Yanting Zhang, Hangyue Chen, and Gaoang Wang. Diffashion: Reference-based fashion design with structure-aware transfer by diffusion models. IEEE Transactions on Multimedia, 2023. 3" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 62, + 273, + 295, + 316 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 273, + 295, + 316 + ], + "spans": [ + { + "bbox": [ + 62, + 273, + 295, + 316 + ], + "type": "text", + "content": "[5] Mengting Chen, Xi Chen, Zhonghua Zhai, Chen Ju, Xuewen Hong, Jinsong Lan, and Shuai Xiao. Wear-any-way: Manipulable virtual try-on via sparse correspondence alignment. arXiv preprint arXiv:2403.12965, 2024. 3" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 62, + 319, + 295, + 373 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 319, + 295, + 373 + ], + "spans": [ + { + "bbox": [ + 62, + 319, + 295, + 373 + ], + "type": "text", + "content": "[6] Seunghwan Choi, Sunghyun Park, Minsoo Lee, and Jaegul Choo. Viton-hd: High-resolution virtual try-on via misalignment-aware normalization. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14131-14140, 2021. 6" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 62, + 375, + 295, + 407 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 375, + 295, + 407 + ], + "spans": [ + { + "bbox": [ + 62, + 375, + 295, + 407 + ], + "type": "text", + "content": "[7] Yisol Choi, Sangkyung Kwak, Kyungmin Lee, Hyungwon Choi, and Jinwoo Shin. Improving diffusion models for virtual try-on. ECCV, 2024. 3, 6" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 62, + 409, + 295, + 463 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 409, + 295, + 463 + ], + "spans": [ + { + "bbox": [ + 62, + 409, + 295, + 463 + ], + "type": "text", + "content": "[8] Zheng Chong, Xiao Dong, Haoxiang Li, Shiyue Zhang, Wenqing Zhang, Xujie Zhang, Hanqing Zhao, and Xiaodan Liang. Catvton: Concatenation is all you need for virtual try-on with diffusion models. arXiv preprint arXiv:2407.15886, 2024. 1" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 62, + 466, + 295, + 520 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 62, + 466, + 295, + 520 + ], + "spans": [ + { + "bbox": [ + 62, + 466, + 295, + 520 + ], + "type": "text", + "content": "[9] Morelli Davide, Fincato Matteo, Cornia Marcella, Landi Federico, Cesari Fabio, and Cucchiara Rita. Dress code: High-resolution multi-category virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022. 1" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 57, + 522, + 295, + 576 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 522, + 295, + 576 + ], + "spans": [ + { + "bbox": [ + 57, + 522, + 295, + 576 + ], + "type": "text", + "content": "[10] Haoye Dong, Xiaodan Liang, Xiaohui Shen, Bochao Wang, Hanjiang Lai, Jia Zhu, Zhiting Hu, and Jian Yin. Towards multi-pose guided virtual try-on network. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9026-9035, 2019." + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 57, + 578, + 295, + 632 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 578, + 295, + 632 + ], + "spans": [ + { + "bbox": [ + 57, + 578, + 295, + 632 + ], + "type": "text", + "content": "[11] Chongjian Ge, Yibing Song, Yuying Ge, Han Yang, Wei Liu, and Ping Luo. Disentangled cycle consistency for highly-realistic virtual try-on. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16928-16937, 2021. 1" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 57, + 635, + 295, + 689 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 635, + 295, + 689 + ], + "spans": [ + { + "bbox": [ + 57, + 635, + 295, + 689 + ], + "type": "text", + "content": "[12] Yuying Ge, Yibing Song, Ruimao Zhang, Chongjian Ge, Wei Liu, and Ping Luo. Parser-free virtual try-on via distilling appearance flows. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8485-8493, 2021. 3" + } + ] + } + ], + "index": 12 + }, + { + "bbox": [ + 57, + 691, + 295, + 713 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 57, + 691, + 295, + 713 + ], + "spans": [ + { + "bbox": [ + 57, + 691, + 295, + 713 + ], + "type": "text", + "content": "[13] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and" + } + ] + } + ], + "index": 13 + } + ], + "sub_type": "ref_text" + }, + { + "bbox": [ + 317, + 73, + 553, + 713 + ], + "type": "list", + "angle": 0, + "index": 28, + "blocks": [ + { + "bbox": [ + 335, + 73, + 553, + 95 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 335, + 73, + 553, + 95 + ], + "spans": [ + { + "bbox": [ + 335, + 73, + 553, + 95 + ], + "type": "text", + "content": "Yoshua Bengio. Generative adversarial networks. Communications of the ACM, 63(11):139-144, 2020. 1, 3" + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 317, + 97, + 553, + 151 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 97, + 553, + 151 + ], + "spans": [ + { + "bbox": [ + 317, + 97, + 553, + 151 + ], + "type": "text", + "content": "[14] Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang. Taming the power of diffusion models for high-quality virtual try-on with appearance flow. In Proceedings of the 31st ACM International Conference on Multimedia, pages 7599-7607, 2023. 3" + } + ] + } + ], + "index": 16 + }, + { + "bbox": [ + 317, + 152, + 553, + 196 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 152, + 553, + 196 + ], + "spans": [ + { + "bbox": [ + 317, + 152, + 553, + 196 + ], + "type": "text", + "content": "[15] Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S Davis. Viton: An image-based virtual try-on network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7543-7552, 2018. 1" + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 317, + 198, + 553, + 241 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 198, + 553, + 241 + ], + "spans": [ + { + "bbox": [ + 317, + 198, + 553, + 241 + ], + "type": "text", + "content": "[16] Xintong Han, Xiaojun Hu, Weilin Huang, and Matthew R Scott. Clothflow: A flow-based model for clothed person generation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 10471-10480, 2019." + } + ] + } + ], + "index": 18 + }, + { + "bbox": [ + 317, + 243, + 553, + 286 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 243, + 553, + 286 + ], + "spans": [ + { + "bbox": [ + 317, + 243, + 553, + 286 + ], + "type": "text", + "content": "[17] Sen He, Yi-Zhe Song, and Tao Xiang. Style-based global appearance flow for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3470-3479, 2022. 1" + } + ] + } + ], + "index": 19 + }, + { + "bbox": [ + 317, + 288, + 553, + 342 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 288, + 553, + 342 + ], + "spans": [ + { + "bbox": [ + 317, + 288, + 553, + 342 + ], + "type": "text", + "content": "[18] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems, 30, 2017. 6" + } + ] + } + ], + "index": 20 + }, + { + "bbox": [ + 317, + 344, + 553, + 398 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 344, + 553, + 398 + ], + "spans": [ + { + "bbox": [ + 317, + 344, + 553, + 398 + ], + "type": "text", + "content": "[19] Xiaobin Hu, Xu Peng, Donghao Luo, Xiaozhong Ji, Jinlong Peng, Zhengkai Jiang, Jiangning Zhang, Taisong Jin, Chengjie Wang, and Rongrong Ji. Diffumatting: Synthesizing arbitrary objects with matting-level annotation. arXiv preprint arXiv:2403.06168, 2024. 4" + } + ] + } + ], + "index": 21 + }, + { + "bbox": [ + 317, + 399, + 553, + 454 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 399, + 553, + 454 + ], + "spans": [ + { + "bbox": [ + 317, + 399, + 553, + 454 + ], + "type": "text", + "content": "[20] Catalin Ionescu, Dragos Papava, Vlad Olaru, and Cristian Sminchisescu. Human3. 6m: Large scale datasets and predictive methods for 3d human sensing in natural environments. IEEE transactions on pattern analysis and machine intelligence, 36(7):1325-1339, 2013. 6" + } + ] + } + ], + "index": 22 + }, + { + "bbox": [ + 317, + 456, + 553, + 511 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 456, + 553, + 511 + ], + "spans": [ + { + "bbox": [ + 317, + 456, + 553, + 511 + ], + "type": "text", + "content": "[21] Thibaut Issenhuth, Jérémie Mary, and Clément Calauzenes. Do not mask what you do not need to mask: a parser-free virtual try-on. In Computer Vision-ECCV 2020: 16th European Conference, Glasgow, UK, August 23-28, 2020, Proceedings, Part XX 16, pages 619-635. Springer, 2020. 3" + } + ] + } + ], + "index": 23 + }, + { + "bbox": [ + 317, + 513, + 553, + 567 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 513, + 553, + 567 + ], + "spans": [ + { + "bbox": [ + 317, + 513, + 553, + 567 + ], + "type": "text", + "content": "[22] Johanna Karras, Aleksander Holynski, Ting-Chun Wang, and Ira Kemelmacher-Shlizerman. Dreampose: Fashion image-to-video synthesis via stable diffusion. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 22623-22633. IEEE, 2023. 3" + } + ] + } + ], + "index": 24 + }, + { + "bbox": [ + 317, + 568, + 553, + 623 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 568, + 553, + 623 + ], + "spans": [ + { + "bbox": [ + 317, + 568, + 553, + 623 + ], + "type": "text", + "content": "[23] Bahjat Kawar, Shiran Zada, Oran Lang, Omer Tov, Huiwen Chang, Tali Dekel, Inbar Mosseri, and Michal Irani. Imagic: Text-based real image editing with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6007-6017, 2023. 4" + } + ] + } + ], + "index": 25 + }, + { + "bbox": [ + 317, + 624, + 553, + 679 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 624, + 553, + 679 + ], + "spans": [ + { + "bbox": [ + 317, + 624, + 553, + 679 + ], + "type": "text", + "content": "[24] Jeongho Kim, Guojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. Stableviton: Learning semantic correspondence with latent diffusion model for virtual try-on. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8176-8185, 2024. 3, 6" + } + ] + } + ], + "index": 26 + }, + { + "bbox": [ + 317, + 681, + 553, + 713 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 317, + 681, + 553, + 713 + ], + "spans": [ + { + "bbox": [ + 317, + 681, + 553, + 713 + ], + "type": "text", + "content": "[25] Nupur Kumari, Bingliang Zhang, Richard Zhang, Eli Shechtman, and Jun-Yan Zhu. Multi-concept customization of text-to-image diffusion. In Proceedings of the IEEE/CVF" + } + ] + } + ], + "index": 27 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "text", + "content": "22624" + } + ] + } + ], + "index": 29 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 8 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 72, + 295, + 712 + ], + "type": "list", + "angle": 0, + "index": 13, + "blocks": [ + { + "bbox": [ + 77, + 72, + 294, + 95 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 77, + 72, + 294, + 95 + ], + "spans": [ + { + "bbox": [ + 77, + 72, + 294, + 95 + ], + "type": "text", + "content": "Conference on Computer Vision and Pattern Recognition, pages 1931-1941, 2023. 4" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 98, + 295, + 152 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 98, + 295, + 152 + ], + "spans": [ + { + "bbox": [ + 56, + 98, + 295, + 152 + ], + "type": "text", + "content": "[26] Sangyun Lee, Gyojung Gu, Sunghyun Park, Seunghwan Choi, and Jaegul Choo. High-resolution virtual try-on with misalignment and occlusion-handled conditions. In European Conference on Computer Vision, pages 204-219. Springer, 2022. 1, 3" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 154, + 295, + 198 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 154, + 295, + 198 + ], + "spans": [ + { + "bbox": [ + 56, + 154, + 295, + 198 + ], + "type": "text", + "content": "[27] Yuhan Li, Hao Zhou, Wenxiang Shang, Ran Lin, Xuanhong Chen, and Bingbing Ni. Anyfit: Controllable virtual try-on for any combination of attire across any scenario. arXiv preprint arXiv:2405.18172, 2024. 3" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 56, + 200, + 294, + 222 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 200, + 294, + 222 + ], + "spans": [ + { + "bbox": [ + 56, + 200, + 294, + 222 + ], + "type": "text", + "content": "[28] Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 6" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 56, + 224, + 294, + 267 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 224, + 294, + 267 + ], + "spans": [ + { + "bbox": [ + 56, + 224, + 294, + 267 + ], + "type": "text", + "content": "[29] Wenquan Lu, Yufei Xu, Jing Zhang, Chaoyue Wang, and Dacheng Tao. Handrefiner: Refining malformed hands in generated images by diffusion-based conditional inpainting. In ACM Multimedia 2024, 2023. 4, 7" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 56, + 270, + 294, + 325 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 270, + 294, + 325 + ], + "spans": [ + { + "bbox": [ + 56, + 270, + 294, + 325 + ], + "type": "text", + "content": "[30] Yifang Men, Yiming Mao, Yuning Jiang, Wei-Ying Ma, and Zhouhui Lian. Controllable person image synthesis with attribute-decomposed gan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5084-5093, 2020. 3" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 56, + 327, + 295, + 382 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 327, + 295, + 382 + ], + "spans": [ + { + "bbox": [ + 56, + 327, + 295, + 382 + ], + "type": "text", + "content": "[31] Davide Morelli, Matteo Fincato, Marcella Cornia, Federico Landi, Fabio Cesari, and Rita Cucchiara. Dress code: High-resolution multi-category virtual try-on. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2231-2235, 2022. 6" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 56, + 384, + 294, + 439 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 384, + 294, + 439 + ], + "spans": [ + { + "bbox": [ + 56, + 384, + 294, + 439 + ], + "type": "text", + "content": "[32] Davide Morelli, Alberto Baldrati, Giuseppe Cartella, Marcelia Cornia, Marco Bertini, and Rita Cucchiara. Ladi-vton: Latent diffusion textual-inversion enhanced virtual try-on. In Proceedings of the 31st ACM International Conference on Multimedia, pages 8580-8589, 2023. 3, 6" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 56, + 441, + 294, + 496 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 441, + 294, + 496 + ], + "spans": [ + { + "bbox": [ + 56, + 441, + 294, + 496 + ], + "type": "text", + "content": "[33] Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4296–4304, 2024. 4" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 56, + 498, + 294, + 553 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 498, + 294, + 553 + ], + "spans": [ + { + "bbox": [ + 56, + 498, + 294, + 553 + ], + "type": "text", + "content": "[34] Maxime Oquab, Timothee Darcet, Theo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 5" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 56, + 555, + 295, + 609 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 555, + 295, + 609 + ], + "spans": [ + { + "bbox": [ + 56, + 555, + 295, + 609 + ], + "type": "text", + "content": "[35] Georgios Pavlakos, Dandan Shan, Ilija Radosavovic, Angjoo Kanazawa, David Fouhey, and Jitendra Malik. Reconstructing hands in 3d with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9826-9836, 2024. 4, 5" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 56, + 613, + 294, + 656 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 613, + 294, + 656 + ], + "spans": [ + { + "bbox": [ + 56, + 613, + 294, + 656 + ], + "type": "text", + "content": "[36] Anton Pelykh, Ozge Mercanoglu Sincan, and Richard Bowden. Giving a hand to diffusion models: a two-stage approach to improving conditional human image generation. arXiv preprint arXiv:2403.10731, 2024. 4" + } + ] + } + ], + "index": 11 + }, + { + "bbox": [ + 56, + 658, + 294, + 712 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 658, + 294, + 712 + ], + "spans": [ + { + "bbox": [ + 56, + 658, + 294, + 712 + ], + "type": "text", + "content": "[37] Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952, 2023. 4" + } + ] + } + ], + "index": 12 + } + ], + "sub_type": "ref_text" + }, + { + "bbox": [ + 316, + 73, + 555, + 712 + ], + "type": "list", + "angle": 0, + "index": 26, + "blocks": [ + { + "bbox": [ + 316, + 73, + 553, + 117 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 73, + 553, + 117 + ], + "spans": [ + { + "bbox": [ + 316, + 73, + 553, + 117 + ], + "type": "text", + "content": "[38] Sergey Prokudin, Christoph Lassner, and Javier Romero. Efficient learning on point clouds with basis point sets. In Proceedings of the IEEE/CVF international conference on computer vision, pages 4332-4341, 2019. 5" + } + ] + } + ], + "index": 14 + }, + { + "bbox": [ + 316, + 119, + 555, + 173 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 119, + 555, + 173 + ], + "spans": [ + { + "bbox": [ + 316, + 119, + 555, + 173 + ], + "type": "text", + "content": "[39] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684-10695, 2022. 1, 4, 6" + } + ] + } + ], + "index": 15 + }, + { + "bbox": [ + 316, + 175, + 553, + 239 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 175, + 553, + 239 + ], + "spans": [ + { + "bbox": [ + 316, + 175, + 553, + 239 + ], + "type": "text", + "content": "[40] Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22500-22510, 2023. 4" + } + ] + } + ], + "index": 16 + }, + { + "bbox": [ + 316, + 242, + 553, + 295 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 242, + 553, + 295 + ], + "spans": [ + { + "bbox": [ + 316, + 242, + 553, + 295 + ], + "type": "text", + "content": "[41] Chitwan Sahara, William Chan, Huiwen Chang, Chris Lee, Jonathan Ho, Tim Salimans, David Fleet, and Mohammad Norouzi. Palette: Image-to-image diffusion models. In ACM SIGGRAPH 2022 conference proceedings, pages 1-10, 2022. 4" + } + ] + } + ], + "index": 17 + }, + { + "bbox": [ + 316, + 297, + 553, + 342 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 297, + 553, + 342 + ], + "spans": [ + { + "bbox": [ + 316, + 297, + 553, + 342 + ], + "type": "text", + "content": "[42] Dvir Samuel, Rami Ben-Ari, Simon Raviv, Nir Darshan, and Gal Chechik. Generating images of rare concepts using pretrained diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4695-4703, 2024. 4" + } + ] + } + ], + "index": 18 + }, + { + "bbox": [ + 316, + 343, + 553, + 397 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 343, + 553, + 397 + ], + "spans": [ + { + "bbox": [ + 316, + 343, + 553, + 397 + ], + "type": "text", + "content": "[43] Ke Sun, Jian Cao, Qi Wang, Linrui Tian, Xindi Zhang, Lian Zhuo, Bang Zhang, Liefeng Bo, Wenbo Zhou, Weiming Zhang, et al. Outfitanyone: Ultra-high quality virtual try-on for any clothing and any person. arXiv preprint arXiv:2407.16224, 2024. 3" + } + ] + } + ], + "index": 19 + }, + { + "bbox": [ + 316, + 399, + 553, + 443 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 399, + 553, + 443 + ], + "spans": [ + { + "bbox": [ + 316, + 399, + 553, + 443 + ], + "type": "text", + "content": "[44] Chengrui Wang, Pengfei Liu, Min Zhou, Ming Zeng, Xubin Li, Tiezheng Ge, et al. Rhand: Refining malformed hands for generated images with decoupled structure and style guidance. arXiv preprint arXiv:2404.13984, 2024. 4" + } + ] + } + ], + "index": 20 + }, + { + "bbox": [ + 316, + 445, + 553, + 488 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 445, + 553, + 488 + ], + "spans": [ + { + "bbox": [ + 316, + 445, + 553, + 488 + ], + "type": "text", + "content": "[45] Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600-612, 2004. 6" + } + ] + } + ], + "index": 21 + }, + { + "bbox": [ + 316, + 490, + 553, + 532 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 490, + 553, + 532 + ], + "spans": [ + { + "bbox": [ + 316, + 490, + 553, + 532 + ], + "type": "text", + "content": "[46] Yuxin Wu, Alexander Kirillov, Francisco Massa, Wan-Yen Lo, and Ross Girshick. Detectron2. https://github.com/facebookresearch/detectron2, 2019. 6" + } + ] + } + ], + "index": 22 + }, + { + "bbox": [ + 316, + 535, + 555, + 601 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 535, + 555, + 601 + ], + "spans": [ + { + "bbox": [ + 316, + 535, + 555, + 601 + ], + "type": "text", + "content": "[47] Zhenyu Xie, Zaiyu Huang, Xin Dong, Fuwei Zhao, Haoye Dong, Xijin Zhang, Feida Zhu, and Xiaodan Liang. Gqvton: Towards general purpose virtual try-on via collaborative local-flow global-parsing learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 23550-23559, 2023. 3" + } + ] + } + ], + "index": 23 + }, + { + "bbox": [ + 316, + 602, + 553, + 645 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 602, + 553, + 645 + ], + "spans": [ + { + "bbox": [ + 316, + 602, + 553, + 645 + ], + "type": "text", + "content": "[48] Yuhao Xu, Tao Gu, Weifeng Chen, and Chengcai Chen. Oot-diffusion: Outfitting fusion based latent diffusion for controllable virtual try-on. arXiv preprint arXiv:2403.01779, 2024.6" + } + ] + } + ], + "index": 24 + }, + { + "bbox": [ + 316, + 647, + 553, + 712 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 316, + 647, + 553, + 712 + ], + "spans": [ + { + "bbox": [ + 316, + 647, + 553, + 712 + ], + "type": "text", + "content": "[49] Binxin Yang, Shuyang Gu, Bo Zhang, Ting Zhang, Xuejin Chen, Xiaoyan Sun, Dong Chen, and Fang Wen. Paint by example: Exemplar-based image editing with diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18381-18391, 2023. 3" + } + ] + } + ], + "index": 25 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "text", + "content": "22625" + } + ] + } + ], + "index": 27 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 9 + }, + { + "para_blocks": [ + { + "bbox": [ + 56, + 73, + 294, + 664 + ], + "type": "list", + "angle": 0, + "index": 12, + "blocks": [ + { + "bbox": [ + 56, + 73, + 294, + 127 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 73, + 294, + 127 + ], + "spans": [ + { + "bbox": [ + 56, + 73, + 294, + 127 + ], + "type": "text", + "content": "[50] Han Yang, Ruimao Zhang, Xiaobao Guo, Wei Liu, Wangmeng Zuo, and Ping Luo. Towards photo-realistic virtual try-on by adaptively generating-preserving image content. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7850-7859, 2020. 1" + } + ] + } + ], + "index": 0 + }, + { + "bbox": [ + 56, + 129, + 294, + 172 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 129, + 294, + 172 + ], + "spans": [ + { + "bbox": [ + 56, + 129, + 294, + 172 + ], + "type": "text", + "content": "[51] Zhijing Yang, Junyang Chen, Yukai Shi, Hao Li, Tianshui Chen, and Liang Lin. Occlumix: Towards de-occlusion virtual try-on by semantically-guided mixup. IEEE Transactions on Multimedia, 25:1477–1488, 2023. 3" + } + ] + } + ], + "index": 1 + }, + { + "bbox": [ + 56, + 174, + 294, + 217 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 174, + 294, + 217 + ], + "spans": [ + { + "bbox": [ + 56, + 174, + 294, + 217 + ], + "type": "text", + "content": "[52] Zhendong Yang, Ailing Zeng, Chun Yuan, and Yu Li. Effective whole-body pose estimation with two-stages distillation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4210-4220, 2023. 6" + } + ] + } + ], + "index": 2 + }, + { + "bbox": [ + 56, + 219, + 294, + 261 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 219, + 294, + 261 + ], + "spans": [ + { + "bbox": [ + 56, + 219, + 294, + 261 + ], + "type": "text", + "content": "[53] Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv preprint arXiv:2308.06721, 2023.4" + } + ] + } + ], + "index": 3 + }, + { + "bbox": [ + 56, + 263, + 294, + 318 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 263, + 294, + 318 + ], + "spans": [ + { + "bbox": [ + 56, + 263, + 294, + 318 + ], + "type": "text", + "content": "[54] Jianhao Zeng, Dan Song, Weizhi Nie, Hongshuo Tian, Tongtong Wang, and An-An Liu. Cat-dm: Controllable accelerated virtual try-on with diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8372-8382, 2024. 6" + } + ] + } + ], + "index": 4 + }, + { + "bbox": [ + 56, + 319, + 294, + 362 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 319, + 294, + 362 + ], + "spans": [ + { + "bbox": [ + 56, + 319, + 294, + 362 + ], + "type": "text", + "content": "[55] Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, Jun Zhu, Lionel M Ni, and Heung-Yeung Shum. Dino: Detr with improved denoising anchor boxes for end-to-end object detection. ICLR, 2023. 2" + } + ] + } + ], + "index": 5 + }, + { + "bbox": [ + 56, + 365, + 294, + 397 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 365, + 294, + 397 + ], + "spans": [ + { + "bbox": [ + 56, + 365, + 294, + 397 + ], + "type": "text", + "content": "[56] Haozhuo Zhang, Bin Zhu, Yu Cao, and Yanbin Hao. Hand1000: Generating realistic hands from text with only 1,000 images. arXiv preprint arXiv:2408.15461, 2024. 4" + } + ] + } + ], + "index": 6 + }, + { + "bbox": [ + 56, + 399, + 294, + 441 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 399, + 294, + 441 + ], + "spans": [ + { + "bbox": [ + 56, + 399, + 294, + 441 + ], + "type": "text", + "content": "[57] Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3836-3847, 2023. 4" + } + ] + } + ], + "index": 7 + }, + { + "bbox": [ + 56, + 443, + 294, + 497 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 443, + 294, + 497 + ], + "spans": [ + { + "bbox": [ + 56, + 443, + 294, + 497 + ], + "type": "text", + "content": "[58] Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 586-595, 2018. 6" + } + ] + } + ], + "index": 8 + }, + { + "bbox": [ + 56, + 499, + 294, + 542 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 499, + 294, + 542 + ], + "spans": [ + { + "bbox": [ + 56, + 499, + 294, + 542 + ], + "type": "text", + "content": "[59] Xujie Zhang, Ente Lin, Xiu Li, Yuxuan Luo, Michael Kampffmeyer, Xin Dong, and Xiaodan Liang. Mmtryon: Multi-modal multi-reference control for high-quality fashion generation. arXiv preprint arXiv:2405.00448, 2024. 3" + } + ] + } + ], + "index": 9 + }, + { + "bbox": [ + 56, + 544, + 294, + 597 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 544, + 294, + 597 + ], + "spans": [ + { + "bbox": [ + 56, + 544, + 294, + 597 + ], + "type": "text", + "content": "[60] Shihao Zhao, Dongdong Chen, Yen-Chun Chen, Jianmin Bao, Shaozhe Hao, Lu Yuan, and Kwan-Yee K Wong. Uni-controlnet: All-in-one control to text-to-image diffusion models. Advances in Neural Information Processing Systems, 36, 2024. 4" + } + ] + } + ], + "index": 10 + }, + { + "bbox": [ + 56, + 600, + 294, + 664 + ], + "type": "ref_text", + "angle": 0, + "lines": [ + { + "bbox": [ + 56, + 600, + 294, + 664 + ], + "spans": [ + { + "bbox": [ + 56, + 600, + 294, + 664 + ], + "type": "text", + "content": "[61] Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Sahara, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two unets. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4606-4615, 2023. 1, 2, 3" + } + ] + } + ], + "index": 11 + } + ], + "sub_type": "ref_text" + } + ], + "discarded_blocks": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "page_number", + "angle": 0, + "lines": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "spans": [ + { + "bbox": [ + 293, + 749, + 317, + 757 + ], + "type": "text", + "content": "22626" + } + ] + } + ], + "index": 13 + } + ], + "page_size": [ + 612, + 792 + ], + "page_idx": 10 + } + ], + "_backend": "vlm", + "_version_name": "2.6.4" +} \ No newline at end of file diff --git a/2025/V^2Dial_ Unification of Video and Visual Dialog via Multimodal Experts/a2fbca87-c18a-41dc-864b-f66927e9f633_content_list.json b/2025/V^2Dial_ Unification of Video and Visual Dialog via Multimodal Experts/a2fbca87-c18a-41dc-864b-f66927e9f633_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..5e4e8e0cdcbf16fc3175977e102b0a181d6ac0eb --- /dev/null +++ b/2025/V^2Dial_ Unification of Video and Visual Dialog via Multimodal Experts/a2fbca87-c18a-41dc-864b-f66927e9f633_content_list.json @@ -0,0 +1,1711 @@ +[ + { + "type": "text", + "text": "$\\mathbf{V}^{2}$ Dial: Unification of Video and Visual Dialog via Multimodal Experts", + "text_level": 1, + "bbox": [ + 109, + 125, + 859, + 154 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/eb6eb8aaa27e4b0f18b08931633881735cc23e776a2800be52f4fd8d37365f5c.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 129, + 186, + 834, + 252 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract", + "text_level": 1, + "bbox": [ + 233, + 286, + 313, + 301 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "We present $V^2$ Dial - a novel expert-based model specifically geared towards simultaneously handling image and video input data for multimodal conversational tasks. Current multimodal models primarily focus on simpler tasks (e.g., VQA, VideoQA, video-text retrieval) and often neglect the more challenging conversational counterparts, such as video and visual/image dialog. Moreover, works on both conversational tasks evolved separately from each other despite their apparent similarities, limiting their applicability potential. To this end, we propose to unify both tasks using a single model that for the first time jointly learns the spatial and temporal features of images and videos by routing them through dedicated experts and aligns them using matching and contrastive learning techniques. Furthermore, we systemically study the domain shift between the two tasks by investigating whether and to what extent these seemingly related tasks can mutually benefit from their respective training data. Extensive evaluations on the widely used video and visual dialog datasets of AVSD and VisDial show that our model achieves new state-of-the-art results across four benchmarks both in zero-shot and fine-tuning settings.", + "bbox": [ + 75, + 316, + 473, + 636 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1. Introduction", + "text_level": 1, + "bbox": [ + 76, + 662, + 209, + 679 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Enabled by the availability of large-scale training data [10, 14, 41] and advances in model design [12, 24, 37, 48, 52], the field of vision-and-language learning saw unprecedented success in recent years. However, current multimodal foundational models [7, 32, 37, 50, 54] still mainly focus on single-round tasks (e.g., VQA [8], VideoQA [58], video-text and text-video retrieval [59]). In contrast, the significantly more challenging conversational tasks, such as visual [1, 22] and video dialog [5], received considerably less attention. Furthermore, methods for these different tasks have advanced independently of each other despite the apparent structural similarities between them. They both operate on a visual input (i.e. an image or video), a short visual description (caption), and a dialog history composed of", + "bbox": [ + 75, + 688, + 470, + 900 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/269c233cda7f0815cbdd2c0db7d53c4ec50f90845f5b7802b8cf9e8a9cff3f6c.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 521, + 284, + 874, + 429 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/8856e672e3d2aa85751f93c17eaf2ef9190b679e6e3170682ac2f180819d8ce0.jpg", + "image_caption": [ + "Figure 1. $\\mathbf{V}^2\\mathbf{D}$ ial uses multimodal experts and outperforms state-of-the-art methods on both video and visual dialog in zero-shot and fine-tuning evaluation settings." + ], + "image_footnote": [], + "bbox": [ + 522, + 436, + 872, + 744 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "previous question-answer pairs. On the one hand, visual dialog models [4, 15, 16, 44, 56] have been primarily trained to rank a list of candidate answers using a Next Sentence Prediction (NSP) head similar to BERT [23] and negative sampling. Thus, they are benchmarked using retrieval met", + "bbox": [ + 496, + 824, + 892, + 900 + ], + "page_idx": 0 + }, + { + "type": "header", + "text": "CVF", + "bbox": [ + 106, + 2, + 181, + 42 + ], + "page_idx": 0 + }, + { + "type": "header", + "text": "This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore.", + "bbox": [ + 236, + 0, + 810, + 46 + ], + "page_idx": 0 + }, + { + "type": "page_number", + "text": "8637", + "bbox": [ + 482, + 944, + 514, + 955 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "rics such as recall (R@k) and normalized discounted cumulative gain (NDCG). In contrast, video dialog models [2, 3, 6, 18, 20, 29, 46, 60] are trained to auto-regressively predict the next answer token using teacher forcing [57] and are evaluated using language generation metrics.", + "bbox": [ + 75, + 90, + 468, + 167 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this work, we mitigate the shortcomings of current dialog systems by proposing $\\mathbf{V}^2\\mathbf{Dial}$ - a novel multimodal expert-based model capable of unifying video and visual dialog tasks without any architectural changes. Specifically, we train dedicated multimodal expert layers that separately process the features of each input modality and learn how to align them using matching and contrastive learning techniques. A key novelty of our approach is that we use dedicated experts to jointly learn the spatial and temporal features of images and videos by routing them through the appropriate experts. Then, we couple these layers with a pretrained LLM to align their hidden states. Thanks to its modularity, our model can efficiently tackle image and video input data simultaneously and seamlessly train on both data types. In summary, our contributions are three-fold:", + "bbox": [ + 75, + 167, + 468, + 391 + ], + "page_idx": 1 + }, + { + "type": "list", + "sub_type": "text", + "list_items": [ + "- We propose $\\mathbf{V}^2\\mathbf{Dial}$ - a multimodal expert-based model that unifies visual and video dialog by simultaneously learning from image and video data. As a core novelty, it employs two experts to separately learn the spatial and temporal features of images and videos. $\\mathbf{V}^2\\mathbf{Dial}$ outperforms state-of-the-art models in both zero-shot and fine-tuning settings (see Figure 1).", + "- We are the first to systematically quantify the effect of domain shift between video and visual dialog tasks based on evaluations on the two widely used datasets of AVSD [5] and VisDial [22]. To this end, we propose an alternative ranking scheme that allows computing the VisDial retrieval metric for fully generative models and enables a fair comparison with previous works.", + "- We are the first to evaluate AVSD in a zero-shot setting, which provides a more solid generalization evaluation of video dialog models compared to the fine-tuned setting. For this, we establish the first benchmark comparison of recent state-of-the-art multimodal models." + ], + "bbox": [ + 76, + 393, + 468, + 681 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2. Related Work", + "text_level": 1, + "bbox": [ + 76, + 693, + 218, + 709 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Visual & Video Dialog. Modeled after human-human communication, visual and video dialog involve reasoning about a visual scene in the form of a video or an image through multiple question-answering rounds in natural language. In comparison to their single-round counterparts, VQA [8] and VideoQA [58], dialog models need to additionally reason about the previous dialog history together with the visual grounding and the current question to be able to answer it efficiently. The best performing visual dialog models [4, 42, 56, 61] leverage pre-trained VLMs and are trained using an NSP head, negative sampling, and binary classification loss. At test time, for each question, the can", + "bbox": [ + 75, + 719, + 467, + 900 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "didate answers are ranked based on their respective NSP scores to compute the retrieval metrics. Although some work [15, 16, 55] claim to train generative visual dialog models, they do so by providing a generative mask where each token can only attend to its left tokens. However, they are trained using the NSP head like the discriminative models. However, this training approach is limiting and suboptimal for a unifying model. Thus, we advocate for a fully generative training paradigm and adapt the ranking scheme of VisDial answers to cater to modern generative models.", + "bbox": [ + 496, + 90, + 890, + 241 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In contrast, works on video dialog follow a purely generative training paradigm and achieved great success building on top of powerful pre-trained LLMs [30, 47]. For example, [28, 34] fine-tuned a LLM on AVSD and obtained performance boosts. More recent works [3, 29] combined LLMs with GNNs and pushed the state-of-the-art results even further. Others [60] introduced a regularization loss to mitigate hallucination. Although video dialog emerged as a natural extension to visual dialog with apparent data structure similarities, research on both tasks evolved separately. To this end, we propose a unifying model that can simultaneously learn both tasks without any architectural modifications and for the first time; systemically study the effect of domain shift between both tasks using the AVSD and VisDial v1.0 datasets.", + "bbox": [ + 496, + 243, + 890, + 469 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Multimodal Expert-based Training. Enhancing models with expert-based training has shown promising potential in boosting performance while maintaining computational efficiency [25, 63, 64]. Some works [12, 54] explored using single modality specific experts within a multimodal transformer architecture. Specifically, they used one vision and one language specific FFN after a shared multi-head self-attention block. Other works [35, 43] explored using multiple sparse modality-agnostic experts and trained them using soft-routers. Our work is positioned at the middle ground of the previously mentioned research directions: We propose to use multiple hard-routed experts per modality to be able to capture more fine-grained features compared to a single expert or multiple modality agnostic experts. Specifically, to the best of our knowledge, $\\mathbf{V}^2\\mathbf{Dial}$ is the first model that learns disentangled spatial and temporal features using two dedicated experts that jointly learn from image and video data. In addition, we propose to deploy two separate language experts (for caption and context) in order to tackle the unique challenges of multimodal conversational tasks.", + "bbox": [ + 496, + 484, + 890, + 787 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "3. $\\mathbf{V}^2$ Dial", + "text_level": 1, + "bbox": [ + 500, + 801, + 581, + 820 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "3.1. Joint Problem Formulation", + "text_level": 1, + "bbox": [ + 500, + 830, + 746, + 845 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "We use a fully generative formulation to unify both video and visual dialog tasks. Specifically, given visual input V (video/image), a corresponding visual description (caption", + "bbox": [ + 500, + 854, + 890, + 900 + ], + "page_idx": 1 + }, + { + "type": "page_number", + "text": "8638", + "bbox": [ + 482, + 944, + 514, + 955 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/76d158b2b310140434bcb6f983d2d8afde34438791dbe9bb380d981608e610be.jpg", + "image_caption": [ + "Tr. Stage 1", + "Figure 2. Architectural overview of $\\mathbf{V}^2$ Dial. We adopt a training strategy composed of three stages. First, we only train the multimodal expert layers using spatial-temporal and video/image text matching losses ( $\\mathcal{L}_{\\mathrm{stm}}$ , $\\mathcal{L}_{\\mathrm{vm}}$ ), spatial-temporal and video/image contrastive learning losses ( $\\mathcal{L}_{\\mathrm{stc}}$ , $\\mathcal{L}_{\\mathrm{vtc}}$ ), and masked language modeling loss ( $\\mathcal{L}_{\\mathrm{mlm}}$ ). Second, we couple the expert layers with a frozen pre-trained LLM end-to-end, using a generative loss $\\mathcal{L}_{\\mathrm{gen}}$ to align their hidden representations. Finally, we additionally fine-tune the LLM weights on the downstream benchmarks. Each expert is a feed-forward network (FFN) composed of two fully connected layers." + ], + "image_footnote": [], + "bbox": [ + 84, + 104, + 885, + 330 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "C), a dialog history $\\mathrm{H}_r = \\left\\{\\left(\\mathbb{Q}_1,\\mathbb{A}_1\\right),\\dots,\\left(\\mathbb{Q}_{r - 1},\\mathbb{A}_{r - 1}\\right)\\right\\}$ comprised of the previous question-answer pairs $\\left\\{\\left(\\mathbb{Q}_i,\\mathbb{A}_i\\right)\\right\\}_{i = 1}^{r - 1}$ and the current question $\\mathbb{Q}_r$ , a model is trained to auto-regressively predict a free-form answer $\\mathbb{A}_r$ at round $r$ Specifically, each answer token $\\mathsf{a}_r^i$ satisfies", + "bbox": [ + 75, + 438, + 468, + 516 + ], + "page_idx": 2 + }, + { + "type": "equation", + "text": "\n$$\n\\mathsf {a} _ {r} ^ {i} = \\underset {\\mathsf {a} \\in \\mathcal {V}} {\\arg \\max } \\left[ \\mathbf {p} \\left(\\mathsf {a} | \\mathrm {V}, \\mathrm {C}, \\mathrm {H} _ {r} \\mathsf {Q} _ {r}, \\mathrm {A} _ {r} ^ {< i}\\right) \\right], \\tag {1}\n$$\n", + "text_format": "latex", + "bbox": [ + 138, + 525, + 468, + 554 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\mathbb{A}_r^{