{"id":43406677,"url":"https://github.com/wholebody3d/wholebody3d","last_synced_at":"2026-02-13T19:00:37.095Z","repository":{"id":65752395,"uuid":"547746647","full_name":"wholebody3d/wholebody3d","owner":"wholebody3d","description":"Official repository of Human3.6M 3D WholeBody (H3WB) dataset","archived":false,"fork":false,"pushed_at":"2025-09-23T08:58:38.000Z","size":11493,"stargazers_count":285,"open_issues_count":5,"forks_count":10,"subscribers_count":11,"default_branch":"main","last_synced_at":"2025-09-23T10:29:35.725Z","etag":null,"topics":["3d-human-pose","3d-pose-estimation","3d-whole-body-pose-estimation","deep-learning","face-landmark-detection","foot-estimation","hand-pose-estimation","human-pose-estimation","pose-estimation","pytorch","whole-body","whole-body-pose-estimation"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/wholebody3d.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE.md","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2022-10-08T07:57:58.000Z","updated_at":"2025-09-08T03:04:08.000Z","dependencies_parsed_at":"2025-09-23T10:29:20.672Z","dependency_job_id":null,"html_url":"https://github.com/wholebody3d/wholebody3d","commit_stats":null,"previous_names":[],"tags_count":1,"template":false,"template_full_name":null,"purl":"pkg:github/wholebody3d/wholebody3d","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wholebody3d%2Fwholebody3d","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wholebody3d%2Fwholebody3d/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wholebody3d%2Fwholebody3d/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wholebody3d%2Fwholebody3d/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/wholebody3d","download_url":"https://codeload.github.com/wholebody3d/wholebody3d/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wholebody3d%2Fwholebody3d/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":29414342,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-02-13T06:24:03.484Z","status":"ssl_error","status_checked_at":"2026-02-13T06:23:12.830Z","response_time":78,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.5:443 state=error: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["3d-human-pose","3d-pose-estimation","3d-whole-body-pose-estimation","deep-learning","face-landmark-detection","foot-estimation","hand-pose-estimation","human-pose-estimation","pose-estimation","pytorch","whole-body","whole-body-pose-estimation"],"created_at":"2026-02-02T16:00:34.365Z","updated_at":"2026-02-13T19:00:37.075Z","avatar_url":"https://github.com/wholebody3d.png","language":"Python","funding_links":[],"categories":["🎭 3D Human Pose Estimation"],"sub_categories":["📊 Major 3D Pose Datasets \u0026 Benchmarks"],"readme":"# H3WB: Human3.6M 3D WholeBody Dataset and Benchmark\n\nThis is the official repository for the paper [\"H3WB: Human3.6M 3D WholeBody Dataset and Benchmark\"](https://arxiv.org/abs/2211.15692) (ICCV'23). The repo contains Human3.6M 3D WholeBody (H3WB) annotations proposed in this paper.\n\nFor the 3D whole-body benchmark and results please refer to [benchmark.md](benchmark.md).\n\n## 🆕Updates\n- **`2025/09/25`** [Link to download test set for RGB \u0026rarr; 3D](https://drive.google.com/file/d/1imYpHp_cJS2Q5F7wCTMAf45ZygKQAuDV/view?usp=sharing). \n- **`2024/03/06`** Due to the requests from the community we released the test sets of H3WB dataset. Please refer to [Evaluation](#evaluation) for more information.\n  - [Link to download test set for 2D \u0026rarr; 3D](https://drive.google.com/file/d/1xrzfR63pflnw8L9e631xyKe4TwUOOm4k/view?usp=share_link).\n  - [Link to download test set for I2D \u0026rarr; 3D](https://drive.google.com/file/d/1GBOCjzIW0GLHHLHhGmUcB32Ke8PVjDXH/view?usp=sharing).\n- **`2024/01/09`** H3WB dataset is now supported in [MMPose](https://github.com/open-mmlab/mmpose) 🎉.\n- **`2023/11/21`** We have made H3WB dataset available in a format commonly employed for 3D pose estimation tasks. To facilitate your use of this format, we provide an accompanying data preparation class. We highly recommend this format for your experiments.\n  - [Link to download dataset in the new format](https://drive.google.com/file/d/1LZh4Jsg3_ZKBF0iEPiexzoGHE4srLgfC/view?usp=share_link).\n  - [Link to data preparation class](https://github.com/wholebody3d/wholebody3d/blob/main/utils/prepare_data_h3wb.py).\n\n## Table of Content\n- [About H3WB](#what-is-h3wb)\n- [Dataset](#h3wb-dataset)\n- [Pretrained models](#pretrained-models)\n- [Tasks](#tasks)\n- [Evaluation](#evaluation)\n- [How to cite](#how-to-cite)\n\n## What is H3WB\n\nH3WB is a large-scale dataset for 3D whole-body pose estimation. It is an extension of [Human3.6m dataset](http://vision.imar.ro/human3.6m/) and \ncontains 133 whole-body (17 for body, 6 for feet, 68 for face and 42 for hands) keypoint annotations on 100K images. The skeleton layout is the same as \n[COCO-Wholebody dataset](https://github.com/jin-s13/COCO-WholeBody). Extensions to other popular 3D pose estimation datasets are ongoing and we already \nhave annotations for [Total Capture](http://www.cs.cmu.edu/~hanbyulj/totalcapture/). If you want your favorite multi-view dataset to get whole-body 3D \nannotations, let us know!\n\nExample annotations:\n\n\u003cimg src=\"imgs/1.jpg\" width=\"800\" height=\"400\"\u003e\n\nLayout from COCO-WholeBody: [Image source](https://github.com/jin-s13/COCO-WholeBody).\n\n\u003cimg src=\"imgs/Fig2_anno.png\" width=\"300\" height=\"300\"\u003e\n\n\n## H3WB Dataset\n\n### Download\n\n- Images can be downloaded from the official cite of [Human3.6m dataset](http://vision.imar.ro/human3.6m/).\nWe provide a data preparation [script](datasets/data_preparation.py) to compile Human3.6m videos into images which allows establishing correct correspondence between images and annotations.\n\n- The annotations for H3WB can be downloaded from [here](https://drive.google.com/file/d/1WVscFZcFIxGi_doifFP3GLTsIbU-hjXO/view?usp=sharing) and by default it is put under [datasets/json/](datasets/json/).\n\n- The annotations for T3WB can be downloaded from [here](https://drive.google.com/file/d/155DpAJZ4XY6Mov9pmhrGzRG4PywDlZv2/view?usp=share_link).\n\n- You could also download H3WB dataset in a format commonly employed for 3D pose estimation tasks [here](https://drive.google.com/file/d/1LZh4Jsg3_ZKBF0iEPiexzoGHE4srLgfC/view?usp=share_link). We provide an accompanying [data preparation class](https://github.com/wholebody3d/wholebody3d/blob/main/utils/prepare_data_h3wb.py) for this format. We highly recommend this format for your experiments. The util files camera.py, mocap_dataset.py and skeleton.py are directly taken from [VideoPose](https://github.com/facebookresearch/VideoPose3D/tree/main/common) repository.\n\n\n### Annotation format\nEvery json is in the following structure, but not every json contains all these values. See [Tasks](#Tasks) section.\n```\nXXX.json --- sample id --- 'image_path'\n                        |\n                        -- 'bbox' --- 'x_min'\n                        |          |- 'y_min'\n                        |          |- 'x_max'\n                        |          |- 'y_max'\n                        |\n                        |- 'keypont_2d' --- joint id --- 'x'\n                        |                             |- 'y'\n                        |\n                        |- 'keypont_3d' --- joint id --- 'x'\n                                                      |- 'y'\n                                                      |- 'z'\n                        \n                        \n```\nWe also provide a [script](utils/utils.py) to load json files.\n\n## Pretrained models\n\nH3WB comes with pretrained models that were used to create the datasets. Model implementations can be found in the 'models/' folder. Please find chekpoints in the table below:\n\n| Dataset | Completion | Diffusion Hands | Diffusion Face |\n|---------|------------|-----------------|----------------|\n|  H3WB   | [ckpt](https://drive.google.com/file/d/1eJ-uz6RYtg3emGiu8IAY72UsUbpTVzzK/view?usp=share_link) | [ckpt](https://drive.google.com/file/d/1G17FaOqd5GlSm08wxMaxjwMmRSqgwXeF/view?usp=share_link) | [ckpt](https://drive.google.com/file/d/15ZUUZSqKGzVCNYRo4-PYqfsso7ajIb07/view?usp=share_link) |\n\nPretrained models for the different tasks of the benchmark can be found in [benchmark.md](benchmark.md).\n\n\n## Tasks\n\nWe propose 3 different tasks along with the 3D WholeBody dataset:\n\n#### 2D \u0026rarr; 3D: 2D complete whole-body to 3D complete whole-body lifting\n\n - Use 2Dto3D_train.json for training and validation. It contains 80k 2D and 3D keypoints.\n\n - Use 2Dto3D_test_2d.json for test on leaderboard. It contains 10k 2D keypoints.\n\n#### I2D \u0026rarr; 3D: 2D incomplete whole-body to 3D complete whole-body lifting\n\n - Use 2Dto3D_train.json for training and validation. It contains 80k 2D and 3D keypoints.\n - Please apply masking on yourself during the training. The official masking strategy is as follows:\n    - With 40\\% probability, each keypoint has a 25\\% chance of being masked,\n    - with 20\\% probability, the face is entirely masked,\n    - with 20\\% probability, the left hand is entirely masked,\n    - with 20\\% probability, the right hand is entirely masked.\n\n - Use I2Dto3D_test_2d.json for test on leaderboard. It contains 10k 2D keypoints. Note that this test set is different from the 2Dto3D_test_2d.json.\n\n#### RGB \u0026rarr; 3D: Image to 3D complete whole-body prediction\n\n - Use RGBto3D_train.json for training and validation. It contains 80k image_path, bounding box and 3D keypoints.\n - It has the same samples from the 2Dto3D_train.json, so you can also access to 2D keypoints if needed.\n - Use RGBto3D_test_img.json for test on leaderboard. It contains 20k image_path and bounding box. \n - Note that the test sample ids are not aligned with previous 2 tasks.\n\n## Evaluation\n\n### Validation\nWe do not provide a validation set. We encourage researchers to report 5-fold cross-validation results with average and standard deviation values.\n\n### Evaluation on test set\nWe have released the the test sets of H3WB dataset.\n  - [Link to download test set for 2D \u0026rarr; 3D](https://drive.google.com/file/d/1xrzfR63pflnw8L9e631xyKe4TwUOOm4k/view?usp=share_link).\n  - [Link to download test set for I2D \u0026rarr; 3D](https://drive.google.com/file/d/1GBOCjzIW0GLHHLHhGmUcB32Ke8PVjDXH/view?usp=sharing).\n\nBoth 2D \u0026rarr; 3D and I2D \u0026rarr; 3D test sets contain 10k triplets of {image, 2D coordinates, 3D coordinates}. Note that, in order to prevent cheating on I2D \u0026rarr; 3D task they have different test samples. \n\n### Visualization\nWe provide a [function](utils/utils.py) to visualize 3D whole-body, as well as the evaluation function for the leaderboard in  this [script](test_leaderboard.py). \n\n## Benchmark\n\nPlease refer to [benchmark.md](benchmark.md) for the benchmark results.\n\n## Terms of Use\n\n1. This project is released under the [MIT License](https://github.com/wholebody3d/wholebody3d/blob/main/LICENSE.md). \n\n2. We do not own the copyright of the images. Use of the images must abide by the [Human3.6m License agreement](http://vision.imar.ro/human3.6m/eula.php).\n\n\n## How to cite\n\nIf you find H3WB 3D WholeBody dataset useful for your project, please cite our paper as follows.\n\n\u003e Yue Zhu, Nermin Samet, David Picard, \"H3WB: Human3.6M 3D WholeBody Dataset and benchmark\", ICCV, 2023.\n\nBibTeX entry:\n```\n@InProceedings{Zhu_2023_ICCV,\n    author    = {Zhu, Yue and Samet, Nermin and Picard, David},\n    title     = {H3WB: Human3.6M 3D WholeBody Dataset and Benchmark},\n    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},\n    month     = {October},\n    year      = {2023},\n    pages     = {20166-20177}\n}\n```\n\nPlease also consider citing the following works.\n\n```\n@article{h36m_pami,\n author = {Ionescu, Catalin and Papava, Dragos and Olaru, Vlad and Sminchisescu, Cristian},\n title = {Human3.6M: Large Scale Datasets and Predictive Methods for 3D Human Sensing in Natural Environments},\n journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},\n publisher = {IEEE Computer Society},\n year = {2014}\n} \n \n@inproceedings{IonescuSminchisescu11,\n author = {Catalin Ionescu, Fuxin Li, Cristian Sminchisescu},\n title = {Latent Structured Models for Human Pose Estimation},\n booktitle = {International Conference on Computer Vision},\n year = {2011}\n}\n```\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwholebody3d%2Fwholebody3d","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fwholebody3d%2Fwholebody3d","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwholebody3d%2Fwholebody3d/lists"}