mala-lab/ADShift

How should we partition the dataset, such as the PACS dataset?

Opened this issue · 2 comments

How should we partition the dataset, such as the PACS dataset?
What should be the split ratios, and what random seed to use

You can download the PACS dataset through the link I provided and get the default of partition.

Can you clarify further what do you mean by default partition? Do you mean the text files at PACS/pacs_label/photo_test_kfold.txt, PACS/pacs_label/photo_train_kfold.txt and PACS/pacs_label/photo_crossval_kfold.txt ? The PACS/pacs_label/photo_test_kfold.txt, and PACS/pacs_label/photo_crossval_kfold.txt seem to have a lot of images in common!

Much appreciate detailed response on how to setup the data for PACS to replicate your results