For every dataset, we need to see 100 or so of the images that are being trained on, *as they are being trained on*, so post transforms and more. This task is creating code that does this, and if possible collecting those images.