Pytorch Lightning Loaded Model Gives Different Results Each Time

I have a U-Net architecture and when I train the model and print out some results without loading it, the model gives good results but when I load the model checkpoint file and try to make predictions it predicts random noise, even worse it gives different predictions for each run with the same test set

Here is my code
model.py Untitled in python (nhjznzxc) - PasteCode.io
train.py Untitled in python (umjqprb3) - PasteCode.io
pred.py pastecode .io/s/m9zkpwk5

Note: Since I’m a new user, I’m allowed to put only 2 links on my post so i added spaces to the link for pred.py.

1 Like