Clarification on the division of 20k test images in "generic_test.jpgl"

[XiaoqiWang]

How are the 20k test images in "generic_test.jpgl" divided? I'd like to follow the same method for 5-fold cross-validation.

Different splits seem to affect model performance, and I want to reduce bias in my evaluation. Could you provide details on how the images are partitioned?

[imfing]

Those partitions were created by the original author of the paper. I don’t recall the exact details, but I believe you don’t need to rely on those specific partitions for your work.