Most of the image datasets are exclusively photo-based. (Definitely no anime in ImageNet, MSCOCO, SVHN, CIFAR etc.) I've noticed that many computer vision tools

Most of the image datasets are exclusively photo-based. (Definitely no anime in ImageNet, MSCOCO, SVHN, CIFAR etc.) I've noticed that many computer vision tools seem to do worse on drawings or cartoons. The other day I was putting together a dataset of anime headshots to try to generate them with WGAN, and needed to crop images down to just the faces; the usual OpenCV tools failed utterly, and I had to use an anime-specific face detection library.