{"success":true,"data":{"author":"ESGT Le Mans (original authors); HEIG-VD geomatics group (mirror)","bucketAccessType":"public","bucketName":"heig-vd-geo/3DSES","bucketPrefix":"","category":"Earth Observation & Geospatial","cloudProvider":"huggingface","createdAt":"2026-11-21T20:45:58.896Z","description":"3DSES (3D Segmentation of ESGT point clouds) is a dataset of dense colorized indoor terrestrial laser scanning (TLS) point clouds covering 427 square meters of the ESGT engineering school in Le Mans, France, with dual annotation: point-level semantic labels and a complete 3D CAD model of the building. About 645 million points across 39 scan stations, organized in three tiers: Gold (7 stations with full manual annotation), Silver (27 stations with full annotation) and Bronze (39 stations with pseudo-labels derived from the 3D model only), plus a test_area of 3 held-out stations. Each station is one NumPy .npy array (float64) with X, Y, Z, R, G, B and label columns; the CAD model is a 320 MB OBJ. Suited to indoor point cloud semantic segmentation, scan-to-BIM and learning from model-derived pseudo-labels. This Hugging Face copy is a per-tier mirror kept by the HEIG-VD geomatics group; the authoritative record is Zenodo, DOI 10.5281/zenodo.13323342, licensed CC BY-SA 4.0. Cite the Zenodo record and the 3DSES paper.","displayTitle":"3DSES: Indoor Terrestrial Laser Scanning (TLS) Point Clouds with Semantic Labels and a CAD Building Model, ESGT Le Mans (HEIG-VD mirror)","doi":"10.5281/zenodo.13323342","huggingFaceUrl":"https://huggingface.co/datasets/heig-vd-geo/3DSES","id":"7f56ce43-24b0-5e23-b396-275321efc661","license":"CC-BY-SA-4.0","licenseEvidenceUrl":"https://zenodo.org/api/records/13323342","objectCountEstimate":80,"probeStatus":"ok","referenceUrl":"https://huggingface.co/datasets/heig-vd-geo/3DSES","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"heig-vd-geo/3DSES\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"heig-vd-geo/3DSES\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=heig-vd-geo, license:cc-by-sa-4.0, size_categories:100M<n<1B, downloads=695"],"format":"mixed","generatedAt":"2026-11-21T20:48:48.861Z","layout":{"example":"Bronze/S140.npy","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"934e554d5059010076007b276465736372273a20273c6638272c2027666f727472616e5f6f72646572273a2046616c73652c20277368617065273a2028363937343237322c2038292c207d202020202020202020202020202020202020202020202020202020202020202020202020202020202020202020202020202020200a000000c0c1158d4000000000c8bc804000000060f5075d40000000000060614000000000000061400000000000c05f40000000201212e23f000000000000f03f00000080bf158d40000000e0c9bc804000000020f1075d40000000000060614000000000000061400000000000c05f40000000807070e03f000000000000f03f00000040bd158d40000000e0cbbc804000000000f5075d40000000000060614000000000000061400000000000c05f40000000e0d1d1e13f000000000000f03f00000000bb158d40000000e0cdbc8040000000a0f4075d40000000000060614000000000000061400000000000c05f40000000605151e13f000000000000f03f000000a0b8158d40000000c0cfbc804000000080f8075d40000000000060614000000000002061400000000000405f4000000000f1f0e03f000000000000f03f00000060b6158d40000000c0d1bc80400000006000085d40000000000060614000000000002061400000000000005f40000000807070e03f000000000000f03f","decoded":{"fields":{"descr":"<f8","dtype":"float64","fortranOrder":false,"ndim":2,"shape":"6974272,8"},"format":"npy","summary":"NPY float64 [6974272, 8]"},"encoding":"hex","key":"Bronze/S140.npy","truncatedAt":512}],"sampleObjects":["3DSES_cad_model.obj","Bronze/S140.npy","Bronze/S141.npy","Bronze/S142.npy","Bronze/S143.npy"],"sidecars":[{"bytes":"---\nlicense: cc-by-sa-4.0\ntask_categories:\n  - token-classification\ntags:\n  - point-cloud\n  - lidar\n  - 3d\n  - semantic-segmentation\n  - indoor\n  - terrestrial-laser-scanning\n  - BIM\npretty_name: \"3DSES: Indoor TLS Point Cloud Segmentation\"\nsize_categories:\n  - 100M<n<1B\n---\n\n# 3DSES: Indoor TLS Point Cloud Segmentation Dataset\n\n![Illustration](Illustration.png)\n\nThis repository provides a **mirror of the 3DSES dataset** to facilitate access\nfor research and benchmarking purposes, with per-tier folders for selective downloads.\n\n> **Disclaimer**\n> This dataset is **not owned nor created by the maintainer of this Hugging Face repository**.\n> The **official and authoritative source** is the original Zenodo record:\n> https://zenodo.org/records/13323342\n\n**3DSES** (3D Segmentation of ESGT point clouds) is a dataset of dense indoor\nTerrestrial Laser Scanning (TLS) colorized point clouds covering **427 m²** of an\nengineering school (ESGT, Le Mans, France). It features dual annotation: semantic\nlabels at the point level alongside a complete 3D CAD building model.\n\n## Dataset Summary\n\n| | Stations | Points | Intensity | Manual labels | Pseudo-labels | Columns | Size |\n|---|---|---|---|---|---|---|---|\n| **Gold** | 7 | 44.5M | Yes | Yes | Yes | 9 | 3.0 GB |\n| **Silver** | 27 | 195.5M | Yes | Yes | Yes | 9 | 14 GB |\n| **Bronze** | 39 | 384.9M | Yes | No | Yes | 8 | 23 GB |\n| **test_area** | 3 | 20.7M | No | No | Yes | 7 | 1.1 GB |\n\n**Total: ~645M points across 39 unique scan stations.**\n\n## File Format\n\nEach `.npy` file contains one scan station as a NumPy array (`float64`).\n\n**Gold / Silver (9 columns):**\n\n| Column | Content |\n|--------|---------|\n| 0-2 | X, Y, Z (local survey coordinates) |\n| 3-5 | R, G, B (0-255) |\n| 6 | Intensity (normalized 0-1) |\n| 7 | Manual semantic label |\n| 8 | Pseudo-label (from CAD model alignment) |\n\n**Bronze (8 columns):** same as above without column 7 (no manual labels).\n\n**test_area (7 columns):** X, Y, Z, R, G, B, pseudo-label only.\n\n## Directory Structure\n\n```\nGold/           # 7 stations with full annotations\n  S163.npy ... S179.npy\nSilver/         # 27 stations with full annotations\n  S150.npy ... S179.npy, S295.npy, S296.npy\nBronze/         # 39 stations with pseudo-labels only\n  S140.npy ... S179.npy, S295.npy, S296.npy\ntest_area/      # 3 held-out stations for evaluation\n  S170.npy, S171.npy, S180.npy\n3DSES_cad_model.obj   # 3D CAD building model (320 MB)\nIllustration.png      # Dataset illustration\n```\n\n## Usage\n\n```python\nimport numpy as np\nfrom pathlib import Path\n\n# Load a single station\ndata = np.load(\"Gold/S163.npy\")\nxyz = data[:, :3]          # coordinates\nrgb = data[:, 3:6]         # color (0-255)\nintensity = data[:, 6]     # normalized intensity\nlabels = data[:, 7]        # manual semantic labels\npseudo = data[:, 8]        # pseudo-labels from CAD\n\n# Load with projax3d\nfrom projax3d.opendata import get_provider\nprovider = get_provider(\"3dses\")\nscene = provider.load_scene(\"gold\", output_dir=\"./data/3dses\")\n```\n\n## Tiers\n\n- **Gold**: Stations with the highest annotation quality (manually verified labels\n  + pseudo-labels + intensity). Best for training and evaluation.\n- **Silver**: Larger set with manual labels. Superset of Gold stations.\n- **Bronze**: All stations with pseudo-labels only (automatically generated from\n  CAD model alignment, >95% accuracy). Largest coverage.\n- **test_area**: Held-out stations for benchmarking (pseudo-labels only).\n\n## Citation\n\n```bibtex\n@inproceedings{merizette2025_3dses,\n  title     = {{3DSES}: an indoor Lidar point cloud segmentation dataset\n               with real and pseudo-labels from a {3D} model},\n  author    = {M{\\'e}rizette, Maxime and Audebert, Nicolas and\n               Kervella, Pierre and Verdun, J{\\'e}r{\\^o}me},\n  booktitle = {Proceedings of the 20th International Joint Conference on\n               Computer Vision, Imaging and Computer Graphics Theory and\n               Applications (VISAPP)},\n  year      = {2025},\n  address   = {Porto, Portugal},\n  note      = {arXiv:2501.17534}\n}\n```\n\n## License\n\nThis dataset is distributed under the","encoding":"utf8","key":"README.md","name":"README.md","size":4837,"truncated":true}],"sourceObjectKey":"Bronze/S140.npy","starterCell":"from huggingface_hub import hf_hub_download\npath = hf_hub_download(repo_id=\"heig-vd-geo/3DSES\", filename=\"Bronze/S140.npy\", repo_type=\"dataset\")\nprint(path)","starterSamplePath":"Bronze/S140.npy","totalObjects":80},"summary":"3DSES (3D Segmentation of ESGT point clouds) is a dataset of dense colorized indoor terrestrial laser scanning (TLS) point clouds covering 427 square meters of the ESGT…","title":"3DSES: Indoor Terrestrial Laser Scanning (TLS) Point Clouds with Semantic Labels and a CAD Building Model, ESGT Le Mans (HEIG-VD mirror)","updatedAt":"2026-11-22T00:59:05.060Z","visibility":"public","websiteUrl":"https://doi.org/10.5281/zenodo.13323342"}}