Doesn't Seem to learn or overfit even one image

[I-CANT-CODE]

I am attempting to train the UNET but it doesn't seem to be learning and just spits out garbage values. I even tried to overfit it to a single example but it doesn't even learn that. Ill attach a pdf of the notebook I am using to train. GIT.pdf

[jvanvugt]

Hmm at first sight I can't see any bugs in your code. I haven't trained the model myself in a while, but there is some open source work that uses it. Maybe you could use it to compare to your own code?
https://github.com/lyft/nuscenes-devkit/blob/master/notebooks/Reference%20Model.ipynb http://www.andrewjanowczyk.com/pytorch-unet-for-digital-pathology-segmentation/

[0xzayd]

It's the last layer. Add Sigmoid if you have binary classifciation or Softmax if you have multiple classes