3DMPPE_POSENET_RELEASE

Why we included this project

This is the official PyTorch release of the ICCV 2019 paper on camera distance-aware top-down 3D multi-person pose estimation. You give it a person's bounding box plus an estimate of camera distance, and it recovers the 3D skeleton for each person in a single RGB image. The repo supports training and testing on standard pose datasets like Human3.6M, MPII, and MS COCO, and it includes a quick demo script plus visualization code for inspecting 2D and 3D results. It is a solid fit for researchers reproducing the paper's findings or developers who want a clean reference when prototyping multi-person 3D pose pipelines. Note that it expects a CUDA environment, and since only the PoseNet stage is included, you supply your own person detector to produce the bounding boxes.