I am training a CNN on an image dataset. When I print the classification report, these are the results. I’m not sure why it is not predicting the labels correctly for a few classes. There are sufficient number of samples in each class. Sometimes precision, recall and f1-score become 0.00 for some classes. Any help would be appreciated!
I would try out some common debugging steps from e.g., A Recipe for Training Neural Networks (karpathy.github.io)
such as using a very small training set and also using it as the test set to verify that the model is capable of overfitting a small dataset (e.g., reaches 100% accuracy perfect scores). If that doesn’t happen, then it indicates that there is a bug somewhere in the training loop (e.g., data augmentations not compatible between training/eval, label mismatches, etc.).
Okay thanks will try it out!