prototypical-networks-tf
prototypical-networks-tf copied to clipboard
How to evaluate using CNN?
When I train on my custom dataset. Acc on training data is 95% but when I use backbone CNN to extract image into embedding vector, then I compare distance with cluster, the acc is so low? Is there something wrong with me? Or Should I use different approaches? Thanks!
Hi, did you use the validation subset to monitor accuracy on it?