RTMPose introduces a real-time, top-down human pose estimation framework that jointly optimizes model architecture and training strategy to achieve high accuracy under strict latency constraints.

Original paper: RTMPose: Real-Time Multi-Person Pose Estimation Based on MMPose

RTMPose-M

This model uses the RTMPose-M variant, which strikes a balance between accuracy and inference speed through efficient backbone design and optimized keypoint heads. It is well suited for real-time pose estimation in applications such as human–computer interaction, sports analytics, video surveillance, and edge AI systems.

Model Configuration:

• Reference implementation: Official MMPose RTMPose models
• Original Weight: rtmpose-m_simcc-aic-coco_pt-aic-coco_420e-256x192
• Resolution: 3x256x192
• Support Cooper version:
  • Cooper SDK: [2.5.3.1]
  • Cooper Foundry: [2.2]

Model	Device	compression	Model Link
RTMPose-m	N1-655	Amba_optimized	Model_Link
RTMPose-m	N1-655	Activation_fp16	Model_Link
RTMPose-m	CV7	Amba_optimized	Model_Link
RTMPose-m	CV7	Activation_fp16	Model_Link
RTMPose-m	CV72	Amba_optimized	Model_Link
RTMPose-m	CV72	Activation_fp16	Model_Link
RTMPose-m	CV75	Amba_optimized	Model_Link
RTMPose-m	CV75	Activation_fp16	Model_Link