sebastianjanampa/detrpose

DETRPose is a real-time, end-to-end multi-person pose estimation system for COCO and CrowdPose. Its main workflow configures datasets and a pose model, trains the model, then produces keypoint predictions for evaluation or inference. The model combines a vision backbone, feature encoder, transformer decoder, matching and loss components, and postprocessing. The repository also provides Torch, ONNX, and TensorRT inference/export paths. Some modules are unsampled; edges below are limited to relationships visible in excerpts, the source index, or README documentation.