Upload folder using huggingface_hub

checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a5d906d64f23927c0ab433e67cecc51f86c08787179ea81db7560eb01e86cad
+size 3775448385

condition_embedder.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75c09b31c05ea2d05d0cff6b59bc325fb1ba3c34f7a68e46730584fa9cf482f6
+size 2538205923

galp.yaml

@@ -0,0 +1,130 @@
+module:
+  condition_embedder:
+    backbone:
+      _target_: sam3d_objects.model.backbone.dit.embedder.embedder_fuser.EmbedderFuser
+      embedder_list:
+      - - _target_: sam3d_objects.model.backbone.dit.embedder.dino.Dino
+          dino_model: dinov2_vitl14_reg
+          input_size: 518
+          normalize_images: true
+        - - - rgb_image
+            - full
+      - - _target_: sam3d_objects.model.backbone.dit.embedder.dino.Dino
+          dino_model: dinov2_vitl14_reg
+          input_size: 518
+          normalize_images: true
+        - - - rgb_image_mask
+            - full
+      - - _target_: sam3d_objects.model.backbone.dit.embedder.pointmap.PointPatchEmbed
+          embed_dim: 512
+          input_size: 256
+          patch_size: 8
+          remap_output: linear
+        - - - pointmap
+            - cropped
+          - - rgb_pointmap
+            - full
+      freeze: true
+      projection_net_hidden_dim_multiplier: 4.0
+      use_pos_embedding: learned
+  generator:
+    backbone:
+      _target_: src.sam3d_objects.model.backbone.generator.shortcut.model.ShortCut
+      batch_mode: true
+      cfg_modalities:
+      - shape
+      inference_steps: 2
+      loss_weights:
+        6drotation_normalized: 0.1
+        _target_: src.sam3d_objects.config.utils.make_dict
+        scale: 0.1
+        shape: 0
+        translation: 1.0
+      ratio_cfg_samples_in_self_consistency_target: 0.25
+      rescale_t: 1
+      reverse_fn:
+        _target_: src.sam3d_objects.model.backbone.generator.classifier_free_guidance.ClassifierFreeGuidanceWithExternalUnconditionalProbability
+        backbone:
+          _target_: src.sam3d_objects.model.backbone.tdfy_dit.models.mot_sparse_structure_flow.SparseStructureFlowTdfyWrapper
+          cond_channels: 1024
+          condition_embedder: null
+          force_zeros_cond: true
+          freeze_d_time_embedder: true
+          freeze_shared_parameters: true
+          in_channels: 8
+          is_shortcut_model: false
+          latent_mapping:
+            6drotation_normalized:
+              _target_: src.sam3d_objects.model.backbone.tdfy_dit.models.mm_latent.Latent_out
+              in_channels: 6
+              model_channels: 1024
+              pos_embedder:
+                _target_: src.sam3d_objects.model.backbone.tdfy_dit.models.mm_latent.LearntPositionEmbedder
+                model_channels: 1024
+                token_len: 1
+            scale:
+              _target_: src.sam3d_objects.model.backbone.tdfy_dit.models.mm_latent.Latent_out
+              in_channels: 3
+              model_channels: 1024
+              pos_embedder:
+                _target_: src.sam3d_objects.model.backbone.tdfy_dit.models.mm_latent.LearntPositionEmbedder
+                model_channels: 1024
+                token_len: 1
+            shape:
+              _target_: src.sam3d_objects.model.backbone.tdfy_dit.models.mm_latent.Latent_in
+              in_channels: 8
+              model_channels: 1024
+              pos_embedder:
+                _target_: src.sam3d_objects.model.backbone.tdfy_dit.models.mm_latent.ShapePositionEmbedder
+                model_channels: 1024
+                patch_size: 1
+                resolution: 8
+            translation:
+              _target_: src.sam3d_objects.model.backbone.tdfy_dit.models.mm_latent.Latent_out
+              in_channels: 3
+              model_channels: 1024
+              pos_embedder:
+                _target_: src.sam3d_objects.model.backbone.tdfy_dit.models.mm_latent.LearntPositionEmbedder
+                model_channels: 1024
+                token_len: 1
+            xz2f_rot:
+              _target_: src.sam3d_objects.model.backbone.tdfy_dit.models.mm_latent.Latent_out
+              in_channels: 6
+              model_channels: 1024
+              pos_embedder:
+                _target_: src.sam3d_objects.model.backbone.tdfy_dit.models.mm_latent.LearntPositionEmbedder
+                model_channels: 1024
+                token_len: 1
+          latent_share_transformer:
+            6drotation_normalized:
+            - 6drotation_normalized
+            - translation
+            - scale
+            - xz2f_rot
+          mlp_ratio: 4
+          model_channels: 1024
+          num_blocks: 24
+          num_heads: 16
+          out_channels: 8
+          patch_size: 1
+          pe_mode: ape
+          qk_rms_norm: true
+          resolution: 16
+          use_checkpoint: false
+          use_fp16: false
+        interval:
+        - 0
+        - 500
+        p_unconditional: 0.1
+        strength: 2.0
+        unconditional_handling: add_flag
+      self_consistency_cfg_strength: 2.0
+      self_consistency_prob: 0.25
+      shortcut_loss_weight: 1.0
+      sigma_min: 0.0
+      time_scale: 1000.0
+      training_time_sampler_fn:
+        _partial_: true
+        _target_: src.sam3d_objects.model.backbone.generator.flow_matching.model.lognorm_sampler
+        mean: -1.0
+        std: 1.0

pipeline.yaml

@@ -0,0 +1,102 @@
+_target_: src.sam3d_objects.pipeline.inference_pipeline_pointmap.InferencePipelinePointMap
+ss_generator_config_path: ss_generator.yaml
+ss_generator_ckpt_path: ss_generator.ckpt
+slat_generator_config_path: slat_generator.yaml
+slat_generator_ckpt_path: slat_generator.ckpt
+ss_decoder_config_path: ss_decoder.yaml
+ss_decoder_ckpt_path: ss_decoder.ckpt
+slat_decoder_gs_config_path: slat_decoder_gs.yaml
+slat_decoder_gs_ckpt_path: slat_decoder_gs.ckpt
+slat_decoder_gs_4_config_path: slat_decoder_gs_4.yaml
+slat_decoder_gs_4_ckpt_path: slat_decoder_gs_4.ckpt
+slat_decoder_mesh_config_path: slat_decoder_mesh.yaml
+slat_decoder_mesh_ckpt_path: slat_decoder_mesh.ckpt
+pad_size: 1.0
+dtype: float16
+version: 3dfy_v9
+slat_cfg_strength: 1
+slat_rescale_t: 1
+downsample_ss_dist: 1
+compile_model: true
+ss_condition_input_mapping: []
+ss_preprocessor:
+  _target_: src.sam3d_objects.data.dataset.tdfy.preprocessor.PreProcessor
+  img_mask_joint_transform: []
+  img_mask_pointmap_joint_transform:
+  - _partial_: true
+    _target_: src.sam3d_objects.data.dataset.tdfy.img_and_mask_transforms.resize_all_to_same_size
+  - _partial_: true
+    _target_: src.sam3d_objects.data.dataset.tdfy.img_and_mask_transforms.crop_around_mask_with_padding
+    box_size_factor: 1.2
+    padding_factor: 0.0
+  img_transform:
+    _target_: torchvision.transforms.Compose
+    transforms:
+    - _partial_: true
+      _target_: src.sam3d_objects.data.dataset.tdfy.img_processing.pad_to_square_centered
+    - _target_: torchvision.transforms.Resize
+      size: 518
+  mask_transform:
+    _target_: torchvision.transforms.Compose
+    transforms:
+    - _partial_: true
+      _target_: src.sam3d_objects.data.dataset.tdfy.img_processing.pad_to_square_centered
+    - _target_: torchvision.transforms.Resize
+      interpolation: 0
+      size: 518
+  normalize_pointmap: false
+  pointmap_transform:
+    _target_: torchvision.transforms.Compose
+    transforms:
+    - _partial_: true
+      _target_: src.sam3d_objects.data.dataset.tdfy.img_processing.pad_to_square_centered
+    - _target_: torchvision.transforms.Resize
+      interpolation: 0
+      size: 518
+pose_decoder_name: ScaleShiftInvariant
+depth_model:
+  _target_: src.sam3d_objects.pipeline.depth_models.moge.MoGe
+  model:
+    _target_: moge.model.v1.MoGeModel.from_pretrained
+    pretrained_model_name_or_path: Ruicheng/moge-vitl
+slat_condition_input_mapping: []
+slat_preprocessor:
+  _target_: src.sam3d_objects.data.dataset.tdfy.preprocessor.PreProcessor
+  img_transform:
+    _target_: torchvision.transforms.Compose
+    transforms:
+    - _target_: src.sam3d_objects.data.dataset.tdfy.img_processing.pad_to_square_centered
+      _partial_: true
+    - _target_: torchvision.transforms.Resize
+      size: 518
+  mask_transform:
+    _target_: torchvision.transforms.Compose
+    transforms:
+    - _target_: src.sam3d_objects.data.dataset.tdfy.img_processing.pad_to_square_centered
+      _partial_: true
+    - _target_: torchvision.transforms.Resize
+      size: 518
+      interpolation: 0
+  img_mask_joint_transform:
+  - _target_: src.sam3d_objects.data.dataset.tdfy.img_and_mask_transforms.crop_around_mask_with_padding
+    _partial_: true
+    box_size_factor: 1.2
+    padding_factor: 0.0
+slat_mean:
+- 0.12211431
+- 0.37204156
+- -1.26521907
+- -2.05276058
+- -3.10432536
+- -0.11294304
+- -0.85146744
+- 0.45506954
+slat_std:
+- 2.37326008
+- 2.13174402
+- 2.2413953
+- 2.30589401
+- 2.1191894
+- 1.8969511
+- 2.41684989
+- 2.08374642

ss_enc_conv3d_16l8_fp16.json

@@ -0,0 +1,12 @@
+
+{
+    "name": "SparseStructureEncoder",
+    "args": {
+        "in_channels": 1,
+        "latent_channels": 8,
+        "num_res_blocks": 2,
+        "num_res_blocks_middle": 2,
+        "channels": [32, 128, 512],
+        "use_fp16": true
+    }
+}

ss_enc_conv3d_16l8_fp16.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:107874eeaa0feb82f51b19db5da7db534fb7e7f19e5a122b9ff1bc2e258bfc6d
+size 119068016