Problem with patch and slices when images are not in the CAPS

[ravih18]

When training a model on patches or slices, the CapsDataset instance counts the number of elements per images. For that it needs to access the size of the image. However, if the user launch the extract command with the save_feature option, the pipeline will directly save the patches/slices without saving the images. And then the train command will crash because it cannot find the images.

• We can by default save the images when launching the extract pipeline (even if we extract patch | slice | roi)
• Or we can maybe pass the image size or the number of extracted elements as a meta data through the extract json