The biggest complaint about DeepVariant is that it isn't even a CV problem. I mean you start with a bunch of images, but those get processed and turned into basecalls, which get aligned to the genome, and then turned into an image again (why though?) for DeepVariant. Interesting approach to say the least.
