Did I miss something while reading this. If the dataset is just a bunch of urls, then surely you can find out who is hosting the image and go after them?