RTMPose

RTMPose is a family of real-time pose estimation models from OpenMMLab, distributed inside MMPose, a PyTorch toolbox covering 2D multi-person body pose, hand and facial keypoints, whole-body estimation with 133 keypoints, animal pose, and 3D mesh recovery. The models are engineered for practical deployment, balancing accuracy against latency across multiple sizes, and the line has grown to include RTMO for one-stage real-time multi-person estimation, RTMW whole-body models with flexible input resolutions, and RTMW3D for real-time 3D whole-body pose. MMPose supplies the surrounding infrastructure: a model zoo, training tools with FP16 support, standard benchmarks such as COCO and MPII, and export through MMDeploy to production runtimes. Everything is open source under the Apache 2.0 license. Researchers use the toolbox to develop new methods, while practitioners deploy RTMPose for motion capture, fitness, AR, and animation pipelines.