I have plotted a confusion matrix of my results. It seems like on certain cases the result is poor and this has effected my model performance. How should I do next? Is the dataset poorly distinguishable? Or does my network have more problems?
Your model does not appear to have learned very much about how to identify the different classes. It seems prone to predicting “undamaged” regardless of the real label.
Do you have a cost history plot from training? That will help you know if your model has converged.
Let’s keep this discussion on a single thread.
I’ll close this one, since I just replied on the other thread where you posted these same images.
1 Like