Hi, I am not getting why there are 1 million reference images, where’s there are only 50,000 query and among those also have ground truth for 25,000. Why rest of the reference images are for? since they also have no relation with training set.
Hi gradient.machine – I’d encourage you to read through the Problem Description and 2021 Image Similarity Dataset and Challenge paper linked on the About page, if you haven’t already. Hope this helps.