Which one should I pick for a new project?

I can only help you understand technical details about cmu-perceptual-computing-lab/openpose or dusty-nv/jetson-inference. Recommending which to use for your project depends on your specific needs, which you haven't described.

That said, here's what the RepoPilot analysis and source reveal to help you decide:

cmu-perceptual-computing-lab/openpose:

• Status: Stale — last commit 2 years ago (per RepoPilot verdict)
• Purpose: Real-time multi-person keypoint detection for body, face, hands, and feet (135 keypoints total)
• Architecture: C++, CUDA, CMake-based; uses Caffe deep learning framework for inference
• Use case: Standalone pose estimation library

dusty-nv/jetson-inference:

• Status: Actively maintained (no staleness signal in the data)
• Purpose: General deep learning inference library supporting multiple tasks: image recognition, object detection, segmentation, pose estimation, action recognition, background removal, depth estimation
• Architecture: C++ and Python bindings; supports multiple network types and precision modes
• Use case: Broader inference framework for various vision tasks

Key differences:

• OpenPose is specialized for keypoint detection only; jetson-inference is a general inference framework
• OpenPose hasn't been updated in 2 years; jetson-inference appears actively maintained
• jetson-inference includes pose estimation via poseNet alongside other capabilities

For a new project, consider:

• If you need only multi-person keypoint detection → OpenPose might work but is stale
• If you need multiple vision tasks or active maintenance → jetson-inference is more suitable

Can you clarify what vision task you're targeting?

FOLLOW_UPS:

• What specific vision task do you need?
• Do you need multi-person keypoint detection?
• Are you deploying on Jetson hardware?