Right now when you import images for object detection Digits does a good job of checking that there is a label file for each image. However, after you import many images with many different classes there is no way to check if all classes are actually in the training or validation set. It would be great if Digits could show the list of classes using the default class names, or the custom ones set when importing images, and then counts for how many occurrences there are for each class in the training and validation sets. That way one could make sure that they have not forgotten to include a class, or enough examples of a class in their validation sets. The label shapes shown help, but it would be easier to review and fix if we could see generated lists.
Guest
