I went through each white-on-black image in the dataset and fixed it as much as possible. Though, I am not a biologist, this is most likely the best revision of that dataset publicly available. I also made my annotation tool available. It processes the dataset in train-hand
folder.
Some entirely trashy samples were ripped out of the set. I didn't have time to annotate those from scratch.