[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-93013":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":15,"stars30d":16,"stars90d":15,"forks30d":15,"starsTrendScore":15,"compositeScore":17,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":18,"fork":18,"defaultBranch":19,"hasWiki":20,"hasPages":18,"topics":21,"createdAt":9,"pushedAt":9,"updatedAt":22,"readmeContent":23,"aiSummary":24,"trendingCount":15,"starSnapshotCount":15,"syncStatus":13,"lastSyncTime":25,"discoverSource":26},93013,"proxypose","ruihangzhang97\u002Fproxypose","ruihangzhang97","Official repository for the paper \"ProxyPose: 6-DoF Pose Tracking via Video-to-Video Translation\"",null,"Python",80,4,2,1,0,10,40.1,false,"main",true,[],"2026-07-22 04:02:07","# 🎯 ProxyPose\n\n**ProxyPose: 6-DoF Pose Tracking via Video-to-Video Translation**\n\n[![arXiv](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FarXiv-2607.06555-b31b1b.svg)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06555)\n[![Project Page](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FProject-Page-blue.svg)](https:\u002F\u002Fruihangzhang97.github.io\u002Fproxypose\u002F)\n\n[Ruihang Zhang](https:\u002F\u002Fruihangzhang97.github.io\u002F)\\*\u003Csup>1\u003C\u002Fsup>, [Felix Taubner](https:\u002F\u002Ffelixtaubner.github.io\u002F)\\*\u003Csup>1,2\u003C\u002Fsup>, [Pooja Ravi](https:\u002F\u002F01pooja10.github.io\u002F)\u003Csup>1\u003C\u002Fsup>, [Kiriakos N. Kutulakos](https:\u002F\u002Fwww.cs.toronto.edu\u002F~kyros\u002F)\u003Csup>1,2\u003C\u002Fsup>, [David B. Lindell](https:\u002F\u002Fdavidlindell.com\u002F)\u003Csup>1,2\u003C\u002Fsup>\n\n\u003Csup>1\u003C\u002Fsup>University of Toronto   \u003Csup>2\u003C\u002Fsup>Vector Institute   \\*Equal contribution\n\n---\n\n**TL;DR:** One query pixel in, a full 6‑DoF pose trajectory out.\n\n---\n\n## ⚡️ Quick start\n\n### 🛠️ 1. Install\n\n```bash\n# Clone the repository\ngit clone https:\u002F\u002Fgithub.com\u002Fruihangzhang97\u002Fproxypose.git\ncd proxypose\n\n# Create and activate a conda environment\nconda create -n proxypose python=3.10 -y\nconda activate proxypose\n\n# Install PyTorch \npip install torch torchvision --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu121\n\n# Install PyTorch3D \nexport FORCE_CUDA=1\npip install --no-build-isolation git+https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fpytorch3d.git@stable\n\n# Install ProxyPose and all remaining dependencies\npip install -e .\n```\n\n### 📦 2. Download model weights\n\nThe weights are **downloaded automatically on first run** — no manual steps needed.\n\n| Weight | HuggingFace | Size |\n|--------|-------------|------|\n| Wan2.1-T2V-14B (base) | [`Wan-AI\u002FWan2.1-T2V-14B`](https:\u002F\u002Fhuggingface.co\u002FWan-AI\u002FWan2.1-T2V-14B) | ~30 GB |\n| ProxyPose LoRA | [`ruihangzhang79\u002Fproxypose`](https:\u002F\u002Fhuggingface.co\u002Fruihangzhang79\u002Fproxypose) | ~600 MB |\n\n\n### 🖱️ 3. Pick your prompt point\n\n```bash\nproxypose-annotate --input-video video\u002Fmy_video.mp4\n```\n\nOpens `http:\u002F\u002Flocalhost:7860` in your browser. Click on a query point in the first frame, press **Save**. Coordinates are written to `video\u002Fmy_video.points.json`.\n\n### 🔭 4. (Optional) Estimate focal length with Depth Anything 3\n\nBy default, we assume a 45° horizontal field of view. For improved accuracy, consider using [Depth Anything 3](https:\u002F\u002Fgithub.com\u002FByteDance-Seed\u002FDepth-Anything-3).\n\nFirst, install Depth Anything 3 inside your ProxyPose environment:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FByteDance-Seed\u002FDepth-Anything-3.git\ncd Depth-Anything-3\n\npip install --no-build-isolation -e .\n```\n\nThen, from the ProxyPose repository, run:\n\n```bash\n# Requires:  pip install hatchling editables\npython -m inference.annotation.depth_anything video\u002Fmy_video.mp4\n```\n\n### ✅ 5. Run inference\n\nPass the query JSON saved by the annotator directly to `--prompt`:\n\n```bash\nproxypose-infer \\\n    --video_path   video\u002Fmy_video.mp4 \\\n    --output_path  output\u002Fresult.mp4 \\\n    --prompt       video\u002Fmy_video.points.json \\\n    --depth_anything_path video\u002Fmy_video.da3.npz   # optional, omit to use fixed 45° FOV\n```\n---\n\n## 📖 Citation\n\n```bibtex\n@article{zhang2026proxypose,\n  title={ProxyPose: 6-DoF Pose Tracking via Video-to-Video Translation},\n  author={Ruihang Zhang and Felix Taubner and Pooja Ravi and Kiriakos N. Kutulakos and David B. Lindell},\n  journal={arXiv preprint arXiv:2607.06555},\n  year={2026}\n}\n```","ProxyPose是一个基于视频到视频翻译的6自由度（6-DoF）物体姿态跟踪方法，通过单点交互式标注实现端到端的稠密位姿轨迹估计。其核心采用LoRA微调的大规模视频生成模型（Wan2.1-T2V），将用户在首帧点击的像素点作为查询，联合深度先验（可选Depth Anything 3）输出连续帧中该点对应物体的完整6D位姿序列。技术上融合生成式建模与几何约束，无需训练专用检测器或关键点网络，支持开箱即用的零样本姿态跟踪。适用于AR\u002FVR内容锚定、机器人视觉伺服、工业零件位姿监控等需轻量级、交互式、单点引导的6D跟踪场景。","2026-07-11 02:30:40","CREATED_QUERY"]