[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-72383":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":18,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":22,"hasPages":22,"topics":24,"createdAt":10,"pushedAt":10,"updatedAt":25,"readmeContent":26,"aiSummary":27,"trendingCount":16,"starSnapshotCount":16,"syncStatus":28,"lastSyncTime":29,"discoverSource":30},72383,"sam-3d-body","facebookresearch\u002Fsam-3d-body","facebookresearch","The repository provides code for running inference with the SAM 3D Body Model (3DB), links for downloading the trained model checkpoints and datasets, and example notebooks that show how to use the model.","",null,"Python",3219,382,35,59,0,81,243,312,29.75,"Other",false,"main",[],"2026-06-12 02:03:02","# SAM 3D\n\nSAM 3D Body is one part of SAM 3D, a pair of models for object and human mesh reconstruction. If you’re looking for SAM 3D Objects, [click here](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fsam-3d-objects).\n\n# SAM 3D Body: Robust Full-Body Human Mesh Recovery\n\n\u003Cp align=\"left\">\n\u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.15989\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FarXiv-2602.15989-b31b1b.svg\" alt=\"arXiv\">\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fai.meta.com\u002Fresearch\u002Fpublications\u002Fsam-3d-body-robust-full-body-human-mesh-recovery\u002F\">\u003Cimg src='https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FMeta_AI-Paper-4A90E2?logo=meta&logoColor=white' alt='Paper'>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fai.meta.com\u002Fblog\u002Fsam-3d\u002F\">\u003Cimg src='https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FProject_Page-Blog-9B72F0?logo=googledocs&logoColor=white' alt='Blog'>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Ffacebook\u002Fsam-3d-body-dataset\">\u003Cimg src='https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F🤗_Hugging_Face-Dataset-F59500?logoColor=white' alt='Dataset'>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fwww.aidemos.meta.com\u002Fsegment-anything\u002Feditor\u002Fconvert-body-to-3d\">\u003Cimg src='https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F🤸_Playground-Live_Demo-E85D5D?logoColor=white' alt='Live Demo'>\u003C\u002Fa>\n\u003C\u002Fp>\n\n[Xitong Yang](https:\u002F\u002Fscholar.google.com\u002Fcitations?user=k0qC-7AAAAAJ&hl=en)\\*, [Devansh Kukreja](https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fdevanshkukreja)\\*, [Don Pinkus](https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fdon-pinkus-9140702a)\\*, [Anushka Sagar](https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fanushkasagar), [Taosha Fan](https:\u002F\u002Fscholar.google.com\u002Fcitations?user=3PJeg1wAAAAJ&hl=en), [Jinhyung Park](https:\u002F\u002Fjindapark.github.io\u002F)⚬, [Soyong Shin](https:\u002F\u002Fyohanshin.github.io\u002F)⚬, [Jinkun Cao](https:\u002F\u002Fwww.jinkuncao.com\u002F), [Jiawei Liu](https:\u002F\u002Fjia-wei-liu.github.io\u002F), [Nicolas Ugrinovic](https:\u002F\u002Fwww.iri.upc.edu\u002Fpeople\u002Fnugrinovic\u002F), [Matt Feiszli](https:\u002F\u002Fscholar.google.com\u002Fcitations?user=A-wA73gAAAAJ&hl=en&oi=ao)†, [Jitendra Malik](https:\u002F\u002Fpeople.eecs.berkeley.edu\u002F~malik\u002F)†, [Piotr Dollar](https:\u002F\u002Fpdollar.github.io\u002F)†, [Kris Kitani](https:\u002F\u002Fkriskitani.github.io\u002F)†\n\n***Meta Superintelligence Labs***\n\n*Core Contributor,  ⚬Intern, †Project Lead\n\n![SAM 3D Body Model Architecture](assets\u002Fmodel_diagram.png?raw=true)\n\n**SAM 3D Body (3DB)** is a promptable model for single-image full-body 3D human mesh recovery (HMR). Our method demonstrates state-of-the-art performance, with strong generalization and consistent accuracy in diverse in-the-wild conditions. 3DB estimates the human pose of the body, feet, and hands based on the [Momentum Human Rig](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002FMHR) (MHR), a new parametric mesh representation that decouples skeletal structure and surface shape for improved accuracy and interpretability.\n\n3DB employs an encoder-decoder architecture and supports auxiliary prompts, including 2D keypoints and masks, enabling user-guided inference similar to the SAM family of models. Our model is trained on high-quality annotations from a multi-stage annotation pipeline using differentiable optimization, multi-view geometry, dense keypoint detection, and a data engine to collect and annotated data covering both common and rare poses across a wide range of viewpoints.\n\n## Qualitative Results\n\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth align=\"center\">Input\u003C\u002Fth>\n\u003Cth align=\"center\">\u003Cstrong>SAM 3D Body\u003C\u002Fstrong>\u003C\u002Fth>\n\u003Cth align=\"center\">CameraHMR\u003C\u002Fth>\n\u003Cth align=\"center\">NLF\u003C\u002Fth>\n\u003Cth align=\"center\">HMR2.0b\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\n\u003Ctr>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample1\u002Finput_bbox.png\" alt=\"Sample 1 Input\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample1\u002FSAM 3D Body.png\" alt=\"Sample 1 - SAM 3D Body\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample1\u002Fcamerahmr.png\" alt=\"Sample 1 - CameraHMR\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample1\u002Fnlf.png\" alt=\"Sample 1 - NLF\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample1\u002F4dhumans.png\" alt=\"Sample 1 - 4DHumans (HMR2.0b)\" width=\"160\">\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample2\u002Finput_bbox.png\" alt=\"Sample 2 Input\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample2\u002FSAM 3D Body.png\" alt=\"Sample 2 - SAM 3D Body\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample2\u002Fcamerahmr.png\" alt=\"Sample 2 - CameraHMR\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample2\u002Fnlf.png\" alt=\"Sample 2 - NLF\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample2\u002F4dhumans.png\" alt=\"Sample 2 - 4DHumans (HMR2.0b)\" width=\"160\">\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample3\u002Finput_bbox.png\" alt=\"Sample 3 Input\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample3\u002FSAM 3D Body.png\" alt=\"Sample 3 - SAM 3D Body\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample3\u002Fcamerahmr.png\" alt=\"Sample 3 - CameraHMR\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample3\u002Fnlf.png\" alt=\"Sample 3 - NLF\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample3\u002F4dhumans.png\" alt=\"Sample 3 - 4DHumans (HMR2.0b)\" width=\"160\">\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample4\u002Finput_bbox.png\" alt=\"Sample 4 Input\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample4\u002FSAM 3D Body.png\" alt=\"Sample 4 - SAM 3D Body\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample4\u002Fcamerahmr.png\" alt=\"Sample 4 - CameraHMR\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample4\u002Fnlf.png\" alt=\"Sample 4 - NLF\" width=\"160\">\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Cimg src=\"assets\u002Fqualitative_comparisons\u002Fsample4\u002F4dhumans.png\" alt=\"Sample 4 - 4DHumans (HMR2.0b)\" width=\"160\">\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\n\u003C\u002Ftable>\n\n*Our SAM 3D Body demonstrates superior reconstruction quality with more accurate pose estimation, better shape recovery, and improved handling of occlusions and challenging viewpoints compared to existing approaches.*\n\n## Latest updates\n\n**11\u002F19\u002F2025** -- Checkpoints Launched, Dataset Released, Web Demo and Paper are out!\n\n## Installation\nSee [INSTALL.md](INSTALL.md) for instructions for python environment setup and model checkpoint access.\n\n## Getting Started\n\n3DB can reconstruct 3D full-body human mesh from a single image, optionally with keypoint\u002Fmask prompts and\u002For hand refinement from the hand decoder.\n\nFor a quick start, run our demo script for model inference and visualization with models from [Hugging Face](https:\u002F\u002Fhuggingface.co\u002Ffacebook) (please make sure to follow [INSTALL.md](INSTALL.md) to request access to our checkpoints.).\n\n```bash\n# Download assets from HuggingFace\nhf download facebook\u002Fsam-3d-body-dinov3 --local-dir checkpoints\u002Fsam-3d-body-dinov3\n\n# Run demo script with default ViTdet detector and MoGe2 FOV model\npython demo.py \\\n    --image_folder \u003Cpath_to_images> \\\n    --output_folder \u003Cpath_to_output> \\\n    --checkpoint_path .\u002Fcheckpoints\u002Fsam-3d-body-dinov3\u002Fmodel.ckpt \\\n    --mhr_path .\u002Fcheckpoints\u002Fsam-3d-body-dinov3\u002Fassets\u002Fmhr_model.pt\n\n# To use SAM3 as the detector to align with online playground of SAM3D\npython demo.py \\\n    --image_folder \u003Cpath_to_images> \\\n    --output_folder \u003Cpath_to_output> \\\n    --checkpoint_path .\u002Fcheckpoints\u002Fsam-3d-body-dinov3\u002Fmodel.ckpt \\\n    --mhr_path .\u002Fcheckpoints\u002Fsam-3d-body-dinov3\u002Fassets\u002Fmhr_model.pt \\\n    --detector_name sam3\n```\n\nYou can also try the following lines of code with models loaded directly from [Hugging Face](https:\u002F\u002Fhuggingface.co\u002Ffacebook)\n\n```python\nimport cv2\nimport numpy as np\nfrom notebook.utils import setup_sam_3d_body\nfrom tools.vis_utils import visualize_sample_together\n\n# Set up the estimator\nestimator = setup_sam_3d_body(hf_repo_id=\"facebook\u002Fsam-3d-body-dinov3\")\n\n# Load and process image\nimg_bgr = cv2.imread(\"path\u002Fto\u002Fimage.jpg\")\noutputs = estimator.process_one_image(cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB))\n\n# Visualize and save results\nrend_img = visualize_sample_together(img_bgr, outputs, estimator.faces)\ncv2.imwrite(\"output.jpg\", rend_img.astype(np.uint8))\n```\n\nFor a complete demo with visualization, see [notebook\u002Fdemo_human.ipynb](notebook\u002Fdemo_human.ipynb).\n\n\n## Model Description\n\n### SAM 3D Body checkpoints\n\nThe table below shows the performance of SAM 3D Body checkpoints released on 11\u002F19\u002F2025.\n\n|      **Backbone (size)**       | **3DPW (MPJPE)** |    **EMDB (MPJPE)**     | **RICH (PVE)** | **COCO (PCK@.05)** |  **LSPET (PCK@.05)** | **Freihand (PA-MPJPE)**\n| :------------------: | :----------: | :--------------------: | :-----------------: | :----------------: | :----------------: | :----------------: |\n|  DINOv3-H+ (840M) \u003Cbr \u002F> ([config](https:\u002F\u002Fhuggingface.co\u002Ffacebook\u002Fsam-3d-body-dinov3\u002Fblob\u002Fmain\u002Fmodel_config.yaml), [checkpoint](https:\u002F\u002Fhuggingface.co\u002Ffacebook\u002Fsam-3d-body-dinov3\u002Fblob\u002Fmain\u002Fmodel.ckpt))   |      54.8      |          61.7         |       60.3        |       86.5        | 68.0 | 5.5\n|   ViT-H  (631M) \u003Cbr \u002F> ([config](https:\u002F\u002Fhuggingface.co\u002Ffacebook\u002Fsam-3d-body-vith\u002Fblob\u002Fmain\u002Fmodel_config.yaml), [checkpoint](https:\u002F\u002Fhuggingface.co\u002Ffacebook\u002Fsam-3d-body-vith\u002Fblob\u002Fmain\u002Fmodel.ckpt))    |     54.8   |         62.9         |       61.7        |        86.8       | 68.9 |  5.5\n\n\n## SAM 3D Body Dataset\nThe SAM 3D Body data is released on [Hugging Face](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Ffacebook\u002Fsam-3d-body-dataset). Please follow the [instructions](.\u002Fdata\u002FREADME.md) to download and process the data.\n\n## SAM 3D Objects\n\n[SAM 3D Objects](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fsam-3d-objects) is a foundation model that reconstructs full 3D shape geometry, texture, and layout from a single image.\n\nAs a way to combine the strengths of both **SAM 3D Objects** and **SAM 3D Body**, we provide an example notebook that demonstrates how to combine the results of both models such that they are aligned in the same frame of reference. Check it out [here](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fsam-3d-objects\u002Fblob\u002Fmain\u002Fnotebook\u002Fdemo_3db_mesh_alignment.ipynb).\n\n## License\n\nThe SAM 3D Body model checkpoints and code are licensed under [SAM License](.\u002FLICENSE).\n\n## Contributing\n\nSee [contributing](CONTRIBUTING.md) and the [code of conduct](CODE_OF_CONDUCT.md).\n\n## Contributors\n\nThe SAM 3D Body project was made possible with the help of many contributors:\nVivian Lee, George Orlin, Nikhila Ravi, Andrew Westbury, Jyun-Ting Song, Zejia Weng, Xizi Zhang, Yuting Ye, Federica Bogo, Ronald Mallet, Ahmed Osman, Rawal Khirodkar, Javier Romero, Carsten Stoll, Jean-Charles Bazin, Sofien Bouaziz, Yuan Dong, Su Zhaoen, Fabian Prada, Alexander Richard, Michael Zollhoefer, Roman Rädle, Sasha Mitts, Michelle Chan, Yael Yungster, Azita Shokrpour, Helen Klein, Mallika Malhotra, Ida Cheng, Eva Galper.\n\n## Citing SAM 3D Body\n\nIf you use SAM 3D Body or the SAM 3D Body dataset in your research, please use the following BibTeX entry.\n\n```bibtex\n@article{yang2026sam3dbody,\n  title={SAM 3D Body: Robust Full-Body Human Mesh Recovery},\n  author={Yang, Xitong and Kukreja, Devansh and Pinkus, Don and Sagar, Anushka and Fan, Taosha and Park, Jinhyung and Shin, Soyong and Cao, Jinkun and Liu, Jiawei and Ugrinovic, Nicolas and Feiszli, Matt and Malik, Jitendra and Dollar, Piotr and Kitani, Kris},\n  journal={arXiv preprint arXiv:2602.15989},\n  year={2026}\n}\n```\n","SAM 3D Body 是一个用于单图像全身体三维人体网格恢复的模型。其核心功能包括基于Momentum Human Rig（MHR）估计人体、脚和手的姿态，采用编码器-解码器架构，并支持2D关键点和掩码等辅助提示以实现用户引导的推理。该模型通过多阶段注释流程训练而成，具有在多样化野外条件下强大的泛化能力和一致性准确性。适用于需要高精度三维人体重建的应用场景，如虚拟现实、增强现实、运动分析以及人机交互等领域。",2,"2026-06-11 03:41:35","high_star"]