Dataset questions

[LAW1223]

Hi, great work! I'm trying to use the dataset. How many paired samples are there? Currently, the train_cp and ref files don't seem to match up. Regarding paired samples, do you have any suggestions?

[Doby-Xu]

Thank you for your attention!

There are no strictly defined “pairs.”
Each sample in train_cp contains a key named name, which corresponds to a unique identity.
Using this key as an index, you can locate a set of reference images (ranging from a few to several hundred) with the same identity in the ref directory.

To construct paired samples, simply sample one or more images randomly from the corresponding reference folder and pair them with the current train_cp sample.

If anything remains unclear, feel free to continue the discussion~

[LAW1223]

Okay, thanks a lot!