IDEA-Research/DWPose
A two-stage distilled deep learning model for whole-body human pose estimation across multiple model sizes.
Velocity · 7d
+2.6
★ / day
Trend
→steady
star history
DWPose provides effective whole-body pose estimation by leveraging two-stage knowledge distillation to train models of varying sizes for detecting body, hand, and face keypoints. The project is built on MMPose and integrates with ControlNet for improved pose-to-image generation. It serves as a drop-in replacement for OpenPose in generative AI pipelines.