Open danielemolino opened 2 months ago
My guesses so far:
Another thing you can do is just process the entire dataset and then select the test samples from that. Then you can use the default MIMIC metadata files to load the dataloader.
Yes in the end I was able to solve the problem, but still thank you for the answer!
One other question, are the models trained only on Frontal Views?
Yes the models were only trained on Frontal views (PA, AP)
Hi! Thank you so much for your amazing works.
I need to use your trained model for a project. In order to understand how to use the code, I've tried to evaluate the weights densenet121-res224-all on the official test split of the Mimic CXR-JPEG.
But I'm getting terrible performances, so I'm pretty sure I'm doing something wrong, but I can't find out what it is.
Here is an example of how I'm making the predictions:
I get for every class (the one which are part of the Mimic Split) an Auc of almost 0.5. Any idea on there is the mistake?