Machine Learning

Hey guys,

I have been experimenting with self-supervised visual learning a bit. Until now I have only ever used U-Nets and related architectures.

No matter what specific task, images or other parameters I changed I always encountered these stains on my output-images (here marked with green), although sometimes more, sometimes less.

Now I wondered if anybody could tell me where they came from and how I could prevent them?

In the attached picture the input (left) and target (right) are the same, so that I can be sure these stains do not come from a badly designed learning task, yet they still appear (output is the middle image).

Thanks in advance and all the best :D

Edit: added line breaks