{"skill":{"slug":"image-3d-scene-reconstruction","displayName":"Image 3D Scene Reconstruction | 图像3D场景重建","summary":"Reconstruct 3D scenes from single images using depth estimation. 从单张图片重建3D场景结构（深度图、点云、Mesh）。","description":"---\nname: Image 3D Scene Reconstruction | 图像3D场景重建\ndescription: Reconstruct 3D scenes from single images using depth estimation. 从单张图片重建3D场景结构（深度图、点云、Mesh）。\n---\n\n# Image 3D Scene Reconstruction | 图像3D场景重建\n\n从卫星图、航拍图或普通照片重建三维场景结构。基于 DA3Metric-Large（Depth Anything 3）深度估计模型，单张图片即可输出深度图、点云和 3D 模型。\n\nReconstruct 3D scenes from satellite, aerial, or regular photos. Based on DA3Metric-Large (Depth Anything 3), outputs depth maps, point clouds, and 3D models from a single image.\n\n---\n\n## 能力 | Capabilities\n\n- **单图深度估计**：输入一张图片，输出米制深度图（米为单位）\n- **点云生成**：从深度图反投影生成彩色 3D 点云\n- **3DGS 输出**：模型内置 3D Gaussian Splatting 能力\n- **相机位姿估计**：自动估计相机内外参\n- **多图融合**：支持多张图片输入做场景融合\n\n## 使用方式 | Usage\n\n### 快速开始\n\n```bash\ncd ~/.openclaw/workspace/projects/image-3d-scene-reconstruction\npython3 scripts/reconstruct.py --input photo.jpg --output output/\n```\n\n### Python API\n\n```python\nfrom depth_anything_3.api import DepthAnything3\nimport cv2\n\nmodel = DepthAnything3.from_pretrained('depth-anything/DA3Metric-Large')\nmodel = model.cuda().eval()\n\nimg = cv2.imread('photo.jpg')\npred = model.inference([img])\n\ndepth = pred.depth[0]        # [H, W] 米制深度\nextrinsics = pred.extrinsics  # 相机外参\nintrinsics = pred.intrinsics  # 相机内参\n```\n\n### CLI\n\n```bash\n# 单张图片 → 3D 输出\npython3 -m depth_anything_3.cli image photo.jpg --export-dir output/ --export-format glb\n\n# 多张图片 → 融合场景\npython3 -m depth_anything_3.cli images ./photos/ --export-dir output/\n```\n\n## 依赖 | Dependencies\n\n| 包 | 用途 |\n|---|------|\n| depth-anything-3 | 深度估计 + 3D 重建引擎 |\n| opencv-python | 图像处理 |\n| torch + torchvision | PyTorch 深度学习框架 |\n| open3d | 点云处理（可选） |\n| trimesh | Mesh 处理（可选） |\n\n## 硬件要求 | Hardware\n\n- **GPU**：NVIDIA GPU，6GB+ VRAM（GTX 1060 及以上）\n- **CUDA**：12.1+（PyTorch 2.5+）\n- **CPU 模式**：可用但极慢，仅推荐测试\n\n## 项目文件 | Project Files\n\n详见 `~/.openclaw/workspace/projects/image-3d-scene-reconstruction/README.md`\n","topics":["图像"],"tags":{"latest":"0.0.2"},"stats":{"comments":0,"downloads":380,"installsAllTime":14,"installsCurrent":0,"stars":0,"versions":2},"createdAt":1778054516418,"updatedAt":1778492859420},"latestVersion":{"version":"0.0.2","createdAt":1778123883658,"changelog":"1. Format fix and metadata cleanup. 格式修复与元数据清理。","license":"MIT-0"},"metadata":null,"owner":{"handle":"moroiser","userId":"s1708w2t7w31qf7b0gsy7kamnx83gqa6","displayName":"Morois","image":"https://avatars.githubusercontent.com/u/98001108?v=4"},"moderation":null}