Hello,
I have tested the vq-16 checkpoint on imagenet-val, and the rFID result aligns with what is reported in the paper. However, for vq-8, I obtained a result of 0.98, which differs from the 0.59 mentioned in the paper. I would appreciate any insights or suggestions regarding this discrepancy. Is there an issue with the published checkpoint? The result I obtained from my own training is also around 1.0 .
Thank you!