MedSegDiff
MedSegDiff copied to clipboard
DATASET
Hi did you use REFUGE-2 or REFUGE dataset?
yes. They are all open datasets. Simply google and download.
So you have used REFUGE dataset
Because REFUGE dataset has 1200 images (400 each in training, test, validation sets), and REFUGE-2 has 2000 images (1200 in training, and 400 each in validation). In the paper it is written as you have used 1200 images in total. So I suppose it is REFUGE dataset. And I received results in similar bounds while using your approach on REFUGE dataset.
I have same question about this dataset. I agree with your point that the author use REFUGE dataset (1200 images from 2018 year) rather than REFUGE-2 (2000 images form 2020 year). As far as I know, the REFUGE-2 is not available. Do you reproduce the performance in the paper ? Only use train dataset or using train+val ?