lightweight-human-pose-estimation-3d-demo.pytorch

Why we included this project

This demo estimates 3D human poses in real time from a single camera feed, detecting up to 18 keypoints per person and lifting them into 3D coordinates so you see a full skeleton in space rather than a flat 2D overlay. It builds on the Lightweight OpenPose and Single-Shot Multi-Person 3D Pose Estimation approaches, and comes with a pretrained checkpoint plus a small C++ pose extractor module you compile yourself. The optional OpenVINO backend is the practical selling point: it lets you run inference at usable speeds on a plain CPU, and there's a TensorRT path for Jetson-class edge devices. It's a good starting point if you're prototyping something like fitness tracking or gesture control, though it's a reference implementation rather than a maintained library, so expect to adapt the code yourself.