When I train the network, the training accuracy is 1.00, the loss is 0 from begining. what may cause this problem?


#1

When I train the network, the training accuracy is 1.00, the loss is 0 from begining. what may cause this problem? Any advice will be appreciated.


#2

Hi @hdjsjyl, if you posted a reproducible example I would be able to help you more. Some common cause of 100% accuracy are:

  • Data error in the labels (are you passing all 0s, or all 1s)
  • Data Leakage (for example you are passing the label as one of your feature)
  • Wrong loss/metric calculation

#3

I am trying to fit mask prediction branch in Masr R-CNN to keypoint detection. I think your advice is very good. I will try to solve problem from these aspects. If you know Mask R-CNN, we can talk more about that. Thank you very much.