WAPR · Wide-Angle Pose Refinement

WAPR estimates and refines unseen-object 6D poses from RGB-D observations and a metric object mesh.

WAPR 根据 RGB-D 观测与米制物体网格，估计并修正未见物体的 6D 位姿。

pip install -U wapr==0.0.4
python -m wapr.bootstrap

Use a compatible Windows or Linux NVIDIA GPU environment with CUDA-enabled PyTorch. OGL requires a CUDA compiler and a C++ compiler (Visual Studio Build Tools on Windows). Windows without TensorRT uses PyTorch by default; an explicit TensorRT selection requires TensorRT 10. Optional features may require additional dependencies and model access. SAM3D and the verified robot planning stack require Linux; SAM3D weights require your own Hugging Face access or locally supplied files. macOS pose inference is not supported.

请使用兼容的 Windows 或 Linux NVIDIA GPU 环境及支持 CUDA 的 PyTorch。OGL 需要 CUDA 编译器和 C++ 编译器（Windows 使用 Visual Studio Build Tools）。Windows 未安装 TensorRT 时默认使用 PyTorch；显式选择 TensorRT 时需要 TensorRT 10。可选功能可能需要额外依赖和模型访问权限。SAM3D 及已验证的机器人规划环境需要 Linux；SAM3D 权重需使用自己的 Hugging Face 访问权限或本地文件。不支持 macOS 位姿推理。

See the documentation for installation, examples and input requirements, and the project repository for source code.

安装步骤、示例和输入要求见文档，源码见项目仓库。

Code: LGPL-2.1. Model weights: weight license. See third-party notices for additional terms.

代码与模型权重分别遵循上述许可；第三方组件遵循其各自许可。