Hi, I have manage to make a u-net that segments the pancreas and now I want to test it in new unlabeled CTs. I tried to use learn.predict(new CT as Image) but i get a wrong result. i think it must be that that image is not normalized.
I have also tried to normalize it with (x-mean)/std and it doesnt work either. i tried loading an image of the training set and i get diferent results that those i get when i use the same image from the data loader.
Is there a way to test the model in new data?
The answer to this problem is exporting the model with learn.export(“file.pkl”) and then reload in another model with load_learner.
With that learner use predict in the new information.