DWPose is a whole-body human pose estimation framework that uses two-stage knowledge distillation to improve the accuracy and efficiency of pose estimators, enabling detection of body, face, hand, and foot keypoints in a unified model.

Original paper: Effective Whole-body Pose Estimation with Two-stages Distillation

DWPose-t

This model uses the DWPose architecture for whole-body 2D pose estimation, with variants ranging from Tiny to Large to support different accuracy and compute requirements. It is well suited for applications such as human pose tracking, gesture recognition, human–computer interaction, sports analytics, animation, and real-time human-centric perception; DWPose is also used as a pose-conditioning component for image and video generation workflows.

Model Configuration:

• Reference implementation: Official DWPose source code
• Original Weight: DWPose-Tiny
• Dataset: COCO
• Resolution: 3x256x192
• Support Cooper version:
  • Cooper SDK: [2.5.4]
  • Cooper Foundry: [2.3]

Model	Device	compression	Model Link
DWPose-t	N1-655	Activation_fp16	Model_Link
DWPose-t	X7	Activation_fp16	Model_Link
DWPose-t	CV7	Activation_fp16	Model_Link
DWPose-t	CV72	Activation_fp16	Model_Link
DWPose-t	CV75	Activation_fp16	Model_Link