[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-1059":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":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":16,"stars30d":17,"stars90d":15,"forks30d":15,"starsTrendScore":15,"compositeScore":18,"rankGlobal":10,"rankLanguage":10,"license":10,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":19,"hasPages":19,"topics":21,"createdAt":10,"pushedAt":10,"updatedAt":25,"readmeContent":26,"aiSummary":27,"trendingCount":15,"starSnapshotCount":15,"syncStatus":16,"lastSyncTime":28,"discoverSource":29},1059,"Boxer3D","Barath19\u002FBoxer3D","Barath19","AR 3D object detection for iPhone with LiDAR — YOLO 2D + BoxerNet 3D lifting","",null,"Swift",402,46,5,0,2,19,44.92,false,"main",[22,23,24],"3dobjectdetection","arkit","boxer","2026-06-12 04:00:07","# Boxer3D\n\n[![oosmetrics](https:\u002F\u002Fapi.oosmetrics.com\u002Fapi\u002Fv1\u002Fbadge\u002Fachievement\u002Fa689afce-27c9-469c-a0dc-565a9e54d5cf.svg)](https:\u002F\u002Foosmetrics.com\u002Frepo\u002FBarath19\u002FBoxer3D)\n[![oosmetrics](https:\u002F\u002Fapi.oosmetrics.com\u002Fapi\u002Fv1\u002Fbadge\u002Fachievement\u002F91450656-3174-4991-b1bb-3677faae3896.svg)](https:\u002F\u002Foosmetrics.com\u002Frepo\u002FBarath19\u002FBoxer3D)\n[![oosmetrics](https:\u002F\u002Fapi.oosmetrics.com\u002Fapi\u002Fv1\u002Fbadge\u002Fachievement\u002F4f38829f-1e42-4382-8012-bb43fa9e11ec.svg)](https:\u002F\u002Foosmetrics.com\u002Frepo\u002FBarath19\u002FBoxer3D)\n\nAR 3D object detection for iPhone with LiDAR. Detects objects with YOLO and lifts them to 3D oriented bounding boxes using [BoxerNet](https:\u002F\u002Ffacebookresearch.github.io\u002Fboxer\u002F) (Meta Research), displayed in augmented reality.\n\n## Demo\n\n\u003Cimg width=\"1418\" height=\"849\" alt=\"banner\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F05cd68b9-7df5-41d1-95fe-4197fdb539d5\" \u002F>\n\n![Boxer3D](https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fe5c5cbce-bcf3-48d4-9994-d317f647950d)\n\n## How It Works\n\n```\niPhone Camera + LiDAR\n       │\n       ├──► YOLO11n (2D detection) ──► top 3 bounding boxes\n       │\n       ├──► LiDAR depth ──► median depth per 16×16 patch\n       │\n       ├──► ARKit ──► camera pose + intrinsics + gravity\n       │\n       └──► BoxerNet (3D lifting) ──► oriented 3D bounding boxes\n                                           │\n                                     SceneKit AR rendering\n```\n\n1. **YOLO11n** detects objects in 2D (640×640, 80 COCO classes)\n2. **BoxerNet** lifts 2D boxes to 7-DoF 3D boxes (center, size, yaw) using DINOv3 visual features + LiDAR depth\n3. **ARKit** Camera poses + Gravity Vector + LiDAR depth\n4. **SceneKit** renders 3D wireframe boxes anchored in the real world\n\n## Requirements\n\n- iPhone 12 Pro or later (LiDAR required)\n- iOS 16.0+\n- ~450 MB storage for models\n\n## Setup\n\n1. **Clone**\n   ```bash\n   git clone git@github.com:Barath19\u002FBoxer3D.git\n   cd Boxer3D\n   ```\n\n2. **Download models** from [Hugging Face](https:\u002F\u002Fhuggingface.co\u002FBarath\u002Fboxer3d)\n   ```bash\n   pip install huggingface_hub\n   huggingface-cli download Barath\u002Fboxer3d --local-dir boxer\u002F\n   ```\n   This places the following in the `boxer\u002F` directory:\n   - `BoxerNet.onnx` (~391 MB, float32) — exported from BoxerNet checkpoint\n   - `yolo11n.onnx` (~10 MB, float32) — exported from Ultralytics YOLO11n\n\n3. **Open in Xcode**\n   ```bash\n   open boxer.xcodeproj\n   ```\n   Xcode will automatically resolve the ONNX Runtime SPM dependency.\n\n4. **Build & Run** on your iPhone (Cmd+R)\n\n## Models\n\n| Model | Size | Input | Output |\n|-------|------|-------|--------|\n| **yolo11n** | 10 MB | (1, 3, 640, 640) RGB | (1, 84, 8400) boxes + classes |\n| **BoxerNet** | 391 MB | (1, 3, 960, 960) RGB + (1, 1, 60, 60) depth + (1, M, 4) boxes + (1, 3600, 6) rays | (M, 3) center, (M, 3) size, (M,) yaw, (M,) confidence |\n\nBoth models run with ONNX Runtime CoreML Execution Provider for Metal\u002FNeural Engine acceleration.\n\n## Dependencies\n\n- [ONNX Runtime](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fonnxruntime-swift-package-manager) v1.24.2 (via SPM)\n- ARKit, SceneKit, SwiftUI (built-in)\n\n## Roadmap\n\n- [x] Port BoxerNet to Swift\n- [x] Convert BoxerNet.pt to ONNX\n- [x] Upload ONNX weights for download\n- [ ] Optimize for portrait mode\n\n## Acknowledgments\n\nBased on **Boxer: Robust Lifting of Open-World 2D Bounding Boxes to 3D** by Daniel DeTone, Tianwei Shen, Fan Zhang, Lingni Ma, Julian Straub, Richard Newcombe, and Jakob Engel (Meta Reality Labs Research).\n\n- [Project page](https:\u002F\u002Ffacebookresearch.github.io\u002Fboxer\u002F)\n- [GitHub](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fboxer)\n\n```bibtex\n@article{boxer2026,\n  title={Boxer: Robust Lifting of Open-World 2D Bounding Boxes to 3D},\n  author={Daniel DeTone and Tianwei Shen and Fan Zhang and Lingni Ma \n          and Julian Straub and Richard Newcombe and Jakob Engel},\n  year={2026},\n}\n```\n","Boxer3D 是一个基于 iPhone LiDAR 的 AR 3D 物体检测项目。它利用 YOLO 进行 2D 物体检测，并通过 BoxerNet 将检测到的物体提升为 3D 定向边界框，最终在增强现实中显示。该项目的核心功能包括使用 YOLO11n 进行 2D 检测、BoxerNet 进行 3D 提升以及 ARKit 和 SceneKit 进行 AR 渲染。技术特点包括 ONNX Runtime CoreML 执行提供程序支持 Metal\u002FNeural Engine 加速。适合需要在 iPhone 上进行实时 3D 物体检测和增强现实应用的场景，如室内设计、虚拟展示或游戏开发等。要求设备为 iPhone 12 Pro 或更新版本，并运行 iOS 16.0 及以上系统。","2026-06-11 02:41:22","CREATED_QUERY"]