Biased Custom Training #72

I’d like to suggest a feature. Many folks have complained that some of the results “look like a different person” however this makes sense given the existing training data is base on many generic faces.

my idea is to support custom training sets as I suspect if a person were to train with many HD images of a specific face and then try and enhance a blurred or partially obscured image using DFDNet, the results would be significantly better leveraging a custom training set based on the image subject versus the generic training set.