tensorflow_Realtime_Multi-Person_Pose_Estimation icon indicating copy to clipboard operation
tensorflow_Realtime_Multi-Person_Pose_Estimation copied to clipboard
verified publisher icon michalfaber

How to train on custom dataset (no humans)

Open Cyril9227 opened this issue 5 years ago • 0 comments

Hi,

I'm very much interested in using your work for my personal dataset. However I have 2 constraints :

  • Number of keypoints per image is variable (between 2 and 6)
  • Images doesn't represent humans but signals

Any advice on how to train a model on this custom dataset ? I believe I should remove the paf layers and only keep the heatmap...

Cheers

Cyril9227 avatar Mar 02 '21 03:03 Cyril9227