Suggestion - For wrong image classification display a sample image for the wrong class

For wrong image classification when and image is wrongly classified (A is classified as B) display a sample image for the wrong class B. This will allow for the user to get a fell of how wrong is the misclassification. If cat classified as dog --> not too bad. But if classified as motorbyke --> pretty bad.

Picking a random sample image of wrong class B may not always be a good representation of class B. It may also confuse some users thinking this is what network looking at when it is predicting B.

A better way would be for user to browse all the images of selected class to get an idea. We are implementing a dataset browser feature in DLS where user will be able to browse the full dataset , sort, filter and edit values.

Yes having a data browser is better than what I suggested. looking forward to it.