HumanPose

• Extract human pose and better than mmcv or mmdet trash
• The extracted data are in pyskl format
• Including keypoint x, y and confidence
• Easy to use, free of mmcv, mmdet packages

HRnet Usage

• Download HRnet pretrained model pose_hrnet_w32_256*192.pth inBaidu Drive, and move it in ./resource
• Modify the data source path in extract_human_pose.py: which only includes your raw video;
• Modify the output dir in extract_human_pose.py, where the output annotations.pkl is saved;
• Modify the label list in extract_human_pose.py, which is the filename to its ground label, as following;
  • name1.mp4 0
  • name2.mp4 1
  • name3.mp4 2
  • ...
  • nameN.mp4 n
• python extract_human_pose.py --cfg ./config/inference-config.yaml --writeBoxFrames --outputDir ./output TEST.MODEL_FILE ./resource/pose_hrnet_w32_256x192.pth

Mediapipe Usage