SAM3DBody Cpp

SAM3DBody Cpp is a high-performance library for real-time 3D full-body reconstruction from a single RGB camera. It outputs multi-person BVH animations with a 70-joint skeleton, including hands, making it ideal for motion capture, VR, and animation workflows. The pure C++ runtime supports both ONNX and ggml backends, enabling efficient inference on CPU and GPU without Python dependencies. Targeted at developers and researchers, the tool prioritizes speed and low latency, processing video frames on commodity hardware. Its API allows integration into custom applications, while the BVH output is compatible with major animation and game engines. The system uses a neural network model trained on large motion capture datasets to estimate 3D joint positions from 2D images. It handles multiple people in real-time, though performance degrades with more than 2-3 subjects. The pure C++ implementation ensures minimal overhead and cross-platform compatibility. What sets it apart is its focus on real-time performance without requiring specialized hardware like depth sensors. The combination of ONNX and ggml backends gives users flexibility in deployment, from cloud servers to edge devices. However, the project appears to be a personal/experimental release with limited commercial support or documentation.