{ "@context": { "@language": "en", "@vocab": "https://schema.org/", "citeAs": "cr:citeAs", "column": "cr:column", "conformsTo": "dct:conformsTo", "cr": "http://mlcommons.org/croissant/", "rai": "http://mlcommons.org/croissant/RAI/", "prov": "http://www.w3.org/ns/prov#", "data": { "@id": "cr:data", "@type": "@json" }, "dataType": { "@id": "cr:dataType", "@type": "@vocab" }, "dct": "http://purl.org/dc/terms/", "examples": { "@id": "cr:examples", "@type": "@json" }, "extract": "cr:extract", "field": "cr:field", "fileProperty": "cr:fileProperty", "fileObject": "cr:fileObject", "fileSet": "cr:fileSet", "format": "cr:format", "includes": "cr:includes", "isLiveDataset": "cr:isLiveDataset", "jsonPath": "cr:jsonPath", "key": "cr:key", "md5": "cr:md5", "parentField": "cr:parentField", "path": "cr:path", "recordSet": "cr:recordSet", "references": "cr:references", "regex": "cr:regex", "repeated": "cr:repeated", "replace": "cr:replace", "samplingRate": "cr:samplingRate", "sc": "https://schema.org/", "separator": "cr:separator", "source": "cr:source", "subField": "cr:subField", "transform": "cr:transform", "equivalentProperty": "cr:equivalentProperty" }, "@type": "sc:Dataset", "conformsTo": "http://mlcommons.org/croissant/1.0", "name": "CM-EVS", "description": "CM-EVS accompanies the paper 'CM-EVS: Sparse Panoramic RGB-D-Pose Data for Coverage-Oriented Scene Observation'. This metadata describes the publicly hosted release at dataset commit 840e6d203cc0537ad21df746137dd9951c4a60de: 13,631 Blender indoor ERP RGB-depth-pose frames across 374 scene instances, plus source adapters, metadata, and mirrored curator code. The paper reports 36,373 curated indoor frames across 1,275 scenes including locally regenerated HM3D and ScanNet++ data; these paper-wide counts are not the downloadable frame count of this repository. HM3D and ScanNet++ derived frames are not redistributed. Outdoor sources are represented here by adapters rather than hosted RGB-D frames. The RecordSet indexes the released Blender frames through frame_manifest.csv; RGB/depth paths are relative to blender_indoor/. Corresponding pose JSON paths replace panorama_ with pose_ and .png with .json in rgb_path. Assets are individual files, not archives. Per-file checksums are provided in blender_indoor/SHA256SUMS. See LICENSE.md for the mixed-license matrix.", "version": "1.0.2", "license": { "@type": "sc:CreativeWork", "name": "CC-BY-4.0 (primary) + mixed matrix", "url": "https://creativecommons.org/licenses/by/4.0/", "description": "Primary license URL is Creative Commons Attribution 4.0 International for redistributable Blender indoor frames and documentation. The full release is mixed-license: see LICENSE.md in this repository (adapters MIT + upstream EULA/ToS; curator code MIT)." }, "url": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval", "citeAs": "@inproceedings{cmevs2026, title={{CM-EVS}: Sparse Panoramic {RGB-D}-Pose Data for Coverage-Oriented Scene Observation}, author={Liu, Jiale and Li, Jungang and Yu, Jieming and Yu, Xinglin and Dongfang, Zihao and Shi, Keyu and Ding, Zongjian and Zhang, Jiahuan and Bai, Shunwen and Huang, Haoran and Wang, Yurun and Wu, Yanxi and Yu, Ningzhe and Gao, Yudong and Cheng, Mingjun}, booktitle={Advances in Neural Information Processing Systems (NeurIPS), Evaluations and Datasets Track}, year={2026}}", "creator": [ { "@type": "Person", "name": "Jiale Liu", "affiliation": { "@type": "Organization", "name": "Zhejiang University" } }, { "@type": "Person", "name": "Jungang Li", "affiliation": { "@type": "Organization", "name": "The Hong Kong University of Science and Technology (Guangzhou)" } }, { "@type": "Person", "name": "Jieming Yu", "affiliation": { "@type": "Organization", "name": "The Hong Kong University of Science and Technology" } }, { "@type": "Person", "name": "Xinglin Yu", "affiliation": { "@type": "Organization", "name": "Beijing Institute of Technology, Zhuhai" } }, { "@type": "Person", "name": "Zihao Dongfang", "affiliation": { "@type": "Organization", "name": "The Hong Kong University of Science and Technology (Guangzhou)" } }, { "@type": "Person", "name": "Keyu Shi", "affiliation": { "@type": "Organization", "name": "Harbin Institute of Technology" } }, { "@type": "Person", "name": "Zongjian Ding", "affiliation": { "@type": "Organization", "name": "University of Chinese Academy of Sciences" } }, { "@type": "Person", "name": "Jiahuan Zhang", "affiliation": { "@type": "Organization", "name": "Tianjin University" } }, { "@type": "Person", "name": "Shunwen Bai", "affiliation": { "@type": "Organization", "name": "Zhejiang University" } }, { "@type": "Person", "name": "Haoran Huang", "affiliation": { "@type": "Organization", "name": "Zhejiang University" } }, { "@type": "Person", "name": "Yurun Wang", "affiliation": { "@type": "Organization", "name": "Zhejiang University" } }, { "@type": "Person", "name": "Yanxi Wu", "affiliation": { "@type": "Organization", "name": "Zhejiang University" } }, { "@type": "Person", "name": "Ningzhe Yu", "affiliation": { "@type": "Organization", "name": "Zhejiang University" } }, { "@type": "Person", "name": "Yudong Gao", "affiliation": { "@type": "Organization", "name": "The Hong Kong University of Science and Technology" } }, { "@type": "Person", "name": "Mingjun Cheng", "affiliation": { "@type": "Organization", "name": "Zhejiang University" } } ], "datePublished": "2026-05-01", "keywords": [ "panoramic", "equirectangular", "ERP", "RGB-D", "view planning", "fixed-budget", "data-centric", "viewpoint provenance", "indoor scene understanding", "panoramic depth estimation", "novel view synthesis", "world model pretraining" ], "isLiveDataset": false, "distribution": [ { "@type": "cr:FileObject", "@id": "frame-manifest.csv", "name": "frame-manifest.csv", "description": "Per-frame CSV index. rgb_path/depth_path are relative to blender_indoor/ (e.g. scenes/sence_indoor_0001/panorama_0000.png).", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/metadata/frame_manifest.csv", "encodingFormat": "text/csv", "sha256": "c1bfd6220ae0b8155e281c3cb23b78e009437dd0ee208721e216b0a12d31556f" }, { "@type": "cr:FileObject", "@id": "frame-id-mapping.csv", "name": "frame-id-mapping.csv", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/metadata/frame_id_mapping.csv", "encodingFormat": "text/csv", "sha256": "ad824ef9cfe0a623196a6fff4558a55e5ae4f77fc8e841425ea3a4e22aae1132" }, { "@type": "cr:FileObject", "@id": "scene-id-mapping.csv", "name": "scene-id-mapping.csv", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/metadata/scene_id_mapping.csv", "encodingFormat": "text/csv", "sha256": "10e5ba64134fc22e82f27838514852f58c6b2b74c7932cf8ba6a83b960dfbc1c" }, { "@type": "cr:FileObject", "@id": "splits.json", "name": "splits.json", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/metadata/splits.json", "encodingFormat": "application/json", "sha256": "072493c9e533554c0faafb1d3ba766365152f6c5084e2837d29f4d5279c76ab6" }, { "@type": "cr:FileObject", "@id": "blender-source-manifest.json", "name": "blender-source-manifest.json", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/metadata/source_manifest.json", "encodingFormat": "application/json", "sha256": "b60c4375bf290d41fda8fdce6aa648da0aecbd036ed3d8102b480557ae3d7a05" }, { "@type": "cr:FileObject", "@id": "blender-indoor-sha256sums", "name": "blender_indoor/SHA256SUMS", "description": "Per-file SHA256 manifest for released Blender RGB, depth, and pose assets. RGB/depth digests are published manifest values, not independently rehashed by this metadata update.", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/SHA256SUMS", "encodingFormat": "text/plain", "sha256": "013d3691b054b4d66146fa320d20731bed68888cfb6505b8834d4c553ad79da7" }, { "@type": "cr:FileObject", "@id": "license-md", "name": "LICENSE.md", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/LICENSE.md", "encodingFormat": "text/markdown", "sha256": "f314bef9456b170ff9db4a6e7a2b66cb799eb1fd666e0ea5e04d4e6a51c01156" }, { "@type": "cr:FileObject", "@id": "readme-md", "name": "README.md", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/README.md", "encodingFormat": "text/markdown", "sha256": "76053bd0c17c2c726fe9acb9eabcb7508768a0f6970fde02a74da90b8b172612", "description": "Repository README at the pinned snapshot; some historical release prose is superseded by the explicit availability statements in this Croissant." }, { "@type": "cr:FileObject", "@id": "reviewer-sample-pose-0000", "name": "sence_indoor_0001/pose_0000.json", "description": "Reviewer sample frame pose (small file; SHA computed locally). Path spelling is sence_indoor_0001 as on the hub.", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/scenes/sence_indoor_0001/pose_0000.json", "encodingFormat": "application/json", "sha256": "4677ae68b14e6f1499a7435fcdf6cedbd67e23b9ed14548513344af0bfc493d3" }, { "@type": "cr:FileObject", "@id": "reviewer-sample-rgb-0000", "name": "sence_indoor_0001/panorama_0000.png", "description": "Reviewer sample RGB; digest from blender_indoor/SHA256SUMS (not re-downloaded for hashing).", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/scenes/sence_indoor_0001/panorama_0000.png", "encodingFormat": "image/png", "sha256": "0c742d89236b18df47f34753bcca8fe187a9b2e479775fc4e88b80c0ebca857f" }, { "@type": "cr:FileObject", "@id": "reviewer-sample-depth-0000", "name": "sence_indoor_0001/panorama_0000_depth.npy", "description": "Reviewer sample depth; digest from blender_indoor/SHA256SUMS.", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/scenes/sence_indoor_0001/panorama_0000_depth.npy", "encodingFormat": "application/octet-stream", "sha256": "7328f5c6e380101df4dc00fdf825f93d1848783a22a28fb9f49c71ead5747f51" }, { "@type": "cr:FileObject", "@id": "adapters-hm3d-pipeline", "name": "adapters/hm3d/pipeline.py", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/adapters/hm3d/pipeline.py", "encodingFormat": "text/x-python", "sha256": "1008e53f357b0cb3126139b682ceede7d329b6e3dfa2f178b4cf62d7eda591f9" }, { "@type": "cr:FileObject", "@id": "adapters-scannetpp-pipeline", "name": "adapters/scannetpp/pipeline.py", "contentUrl": "https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/adapters/scannetpp/pipeline.py", "encodingFormat": "text/x-python", "sha256": "8c03c3f667f9ee1466d1c183359bf9a4d182473675d07515f8dfcbecfca39928" } ], "recordSet": [ { "@type": "cr:RecordSet", "@id": "erp-frame-records", "name": "erp-frame-records", "description": "One record per publicly released Blender indoor ERP frame: 13,631 records across 374 scene instances. Asset paths are relative to https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/. For each rgb path, the corresponding pose JSON uses pose_ instead of panorama_ and .json instead of .png. The original CSV also contains seven entirely empty reserved columns; these are intentionally omitted from the RecordSet. Pose values are stored in separate JSONs, not inline in this CSV.", "field": [ { "@type": "cr:Field", "@id": "erp-frame-records/frame_id", "name": "frame_id", "dataType": "sc:Text", "source": { "fileObject": { "@id": "frame-manifest.csv" }, "extract": { "column": "frame_id" } }, "description": "Scene-qualified frame identifier." }, { "@type": "cr:Field", "@id": "erp-frame-records/source", "name": "source", "dataType": "sc:Text", "source": { "fileObject": { "@id": "frame-manifest.csv" }, "extract": { "column": "source" } }, "description": "Source label in the released manifest." }, { "@type": "cr:Field", "@id": "erp-frame-records/scene_id", "name": "scene_id", "dataType": "sc:Text", "source": { "fileObject": { "@id": "frame-manifest.csv" }, "extract": { "column": "scene_id" } }, "description": "Released scene-instance identifier." }, { "@type": "cr:Field", "@id": "erp-frame-records/split", "name": "split", "dataType": "sc:Text", "source": { "fileObject": { "@id": "frame-manifest.csv" }, "extract": { "column": "split" } }, "description": "Released split assignment; not a statement of cross-source deduplication." }, { "@type": "cr:Field", "@id": "erp-frame-records/rgb", "name": "rgb", "description": "Relative RGB PNG file path under https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/. This field contains a path string, not decoded image/array bytes.", "dataType": "sc:Text", "source": { "fileObject": { "@id": "frame-manifest.csv" }, "extract": { "column": "rgb_path" } } }, { "@type": "cr:Field", "@id": "erp-frame-records/depth", "name": "depth", "dataType": "sc:Text", "source": { "fileObject": { "@id": "frame-manifest.csv" }, "extract": { "column": "depth_path" } }, "description": "Relative NumPy range-depth file path under https://huggingface.co/datasets/CM-EVS-Dataset/cmevs-erp-eval/resolve/840e6d203cc0537ad21df746137dd9951c4a60de/blender_indoor/. This field contains a path string, not decoded image/array bytes." }, { "@type": "cr:Field", "@id": "erp-frame-records/camera_type", "name": "camera_type", "dataType": "sc:Text", "source": { "fileObject": { "@id": "frame-manifest.csv" }, "extract": { "column": "camera_type" } }, "description": "Manifest value erp_cycles; pose JSONs independently label camera_type as erp_ray." } ] } ], "rai:dataCollection": "The paper describes geometry-aware candidate generation, validity filtering, conflict-aware selection, ERP rendering, and unified export across indoor sources. This hosted snapshot contains only the 374-scene Blender indoor RGB-depth-pose release and adapters/code for other sources. HM3D and ScanNet++ require upstream access and local regeneration; outdoor adapters re-encode source trajectories. Do not interpret paper-wide generated frame counts as redistributed files.", "rai:dataPreprocessingProtocol": "Released RGB files are ERP PNGs and depth files are NumPy arrays. Per-frame pose JSONs contain position, rotation_quaternion, camera_type, coordinate_system, and render_method. All 13,631 downloaded pose JSONs declare 'right-handed, Y-up, Z-forward (cam_to_world)'; they must not be silently interpreted as world-to-camera poses. Follow the stored coordinate_system and explicitly convert conventions when needed. The CSV camera_type is erp_cycles while pose JSON camera_type is erp_ray. This metadata update does not convert any pose or alter frame assets. Relative CSV asset paths are resolved under blender_indoor/ at the pinned release commit.", "rai:dataAnnotationProtocol": "No new manual annotation is asserted for this hosted export. The frame manifest supplies frame_id, source, scene_id, split, rgb_path, depth_path, and camera_type. All 13,631 rows currently have empty room_id, pose_quaternion, pose_position, viewpoint_score, coverage_gain, conflict_ratio, and candidate_id columns; these are not exposed as populated RecordSet fields. Actual pose values are in the separate pose_*.json files. Per-step selection logs and selected_viewpoints.json are not present in this public snapshot, so curator scores and full selection provenance cannot be recovered from this CSV.", "rai:dataReleaseMaintenancePlan": "This Croissant describes a pinned public repository snapshot. Report issues through the dataset repository discussions. Future metadata or asset changes should receive a version/date update and refreshed resource checksums. No fixed future release schedule is asserted.", "rai:dataUseCases": [ "Empirically evaluated in the paper: fixed-budget viewpoint selection using world-voxel coverage under shared candidate pools, and matched ERP novel-view synthesis. These results are limited to the reported protocols, assets, and source splits; this public snapshot is only part of the paper's data.", "Potential applications, not established as validated outcomes by the reported experiments: panoramic depth estimation, Gaussian-splatting reconstruction, and world-model pretraining. Users must establish suitability separately." ], "rai:dataLimitations": [ "Real-scan derived frames (HM3D, ScanNet++) are not redistributed; users must accept upstream license terms and regenerate locally.", "Outdoor frames are re-encoded source trajectories rather than curator-selected subsets and therefore do not carry per-step provenance.", "Synthetic-real transfer must be validated separately by source; we do not claim Blender-only gains imply real-scan gains.", "Geometry-validity filters may fail in atria, semi-outdoor spaces, narrow transitions, noisy scans, or pure point-cloud scenes.", "Only 13,631 Blender indoor frames across 374 scenes are hosted here; paper-wide indoor and outdoor totals include local processing and must not be treated as the public download size.", "CSV pose and curator-score columns are entirely empty in this snapshot; pose JSONs are available separately, but per-step curation logs are not included.", "Exported pose JSONs declare cam_to_world; reconcile this convention explicitly with any world-to-camera convention used by downstream software." ], "rai:personalSensitiveInformation": "No new personal data is collected. Real-scan sources (HM3D, ScanNet++) may depict private indoor layouts and are not redistributed as derived frames. Even regeneration scripts and viewpoint metadata can reveal where observations would be sampled within a private space; users must comply with upstream source access terms.", "rai:dataBiases": [ "Source assets inherit geographic, architectural, and scanning biases.", "HM3D and ScanNet++ are skewed toward scanned residential indoor spaces.", "Blender assets are skewed toward staged residential, office, and architectural scenes.", "Outdoor sources (TartanGround, OB3D) are skewed toward simulator-generated terrain along circular trajectories.", "Synthetic Blender materials may not match real-scan sensor noise." ], "rai:dataSocialImpact": "CM-EVS lowers the engineering cost of producing auditable panoramic RGB-D resources from existing 3D scenes. Positive uses include panoramic perception, data-centric evaluation, view-planning research, and 3D-consistent world-model pretraining. Potential harms include over-trusting synthetic data, obscuring upstream dataset bias, and using real indoor scans in privacy-sensitive settings. The release therefore separates public synthetic frames from licensed real-scan regeneration and documents intended uses, non-uses, and source licenses.", "rai:hasSyntheticData": true, "prov:wasDerivedFrom": [ { "@type": "sc:CreativeWork", "name": "Blender indoor 3D assets (CC0 / CC-BY 4.0 upstream)", "description": "Synthetic indoor scenes rendered to ERP RGB-D-pose; redistributed under CC-BY 4.0 as blender_indoor/." }, { "@type": "sc:CreativeWork", "name": "HM3D", "url": "https://aihabitat.org/datasets/hm3d/", "description": "Adapter-only; no derived frames redistributed." }, { "@type": "sc:CreativeWork", "name": "ScanNet++", "url": "https://kaldir.vc.in.tum.de/scannetpp/", "description": "Adapter-only; no derived frames redistributed." }, { "@type": "sc:CreativeWork", "name": "TartanGround / TartanAir", "description": "Outdoor re-encoding adapter only; upstream license applies to regenerated frames." }, { "@type": "sc:CreativeWork", "name": "OB3D", "description": "Outdoor re-encoding adapter only; upstream license applies to regenerated frames." } ], "prov:wasGeneratedBy": { "@type": "sc:Action", "name": "CM-EVS depth-conflict-aware curator + Cycles ERP render pipeline", "description": "Asset load, coordinate normalization, candidate generation, 26-direction geometric-validity filter, conflict-aware greedy selection, 2048x1024 Cycles ERP rendering, unified schema export. Adapter pipelines re-encode outdoor sources; HM3D/ScanNet++ regeneration is local-only. This describes the paper pipeline; availability of per-step provenance is limited as documented in dataAnnotationProtocol." }, "dateModified": "2026-11-06" }