[R] Observing Stable Results with Transfer Learning

I have encountered an interesting phenomenon while working with StyleGan2ada. Initially, I trained a model, observing that it "diverged". I say "diverged" because the generator seems to use repetitive colors in all synthetics, nothing special, but it is noticeable when I generate large quantities of images. To investigate this further, I decided to train a new model using a previous snapshot of the "diverged" model as a starting point. However, to my surprise, the new model, trained with transfer learning, …

colors generator images machinelearning transfer transfer learning