This seems like a crazy overfitting problem, but the loss isn’t even close to 0 yet.
I am new to Machine Learning and this is my second Neural Network.
What I am trying to do is to convert a grayscale image into an RGB image (doesn’t have to be perfect at all, I just want to see what the network comes up with for images it has never seen before, failure to replicate the original colors is the main interest of mine in this project).
The thing is, my current problem isn’t fun at all, because I get the same mash of colors as output for all the images I input into the network.
This is my network:
The structure of the network itself is defined in ./classes. The implementation is in ./app.py. You can test it yourself if you want by putting images in the proper folders. ./guesses/set should contain testing set images, ./rooms/set should contain input images.
I would be happy if this network simply output different things for different input images, but it doesn’t do that.
Could you help me out with this problem? I would be grateful.