CAT-KD icon indicating copy to clipboard operation
CAT-KD copied to clipboard

Regarding the Reproducibility of Results

Open xhdr0618 opened this issue 1 year ago • 1 comments

Thank you very much for your insightful work, but I found that the results of res324_res84 during the reproduction process do not meet the benchmarks in the paper. Why is that? The result on CIFair-100 is 75.01

xhdr0618 avatar Aug 24 '24 06:08 xhdr0618

Sorry for the late response. Could you check if the hyperparameters were changed and whether the CIFAR-100 experiments were run on a single GPU? Also, is this result averaged over five runs?

GzyAftermath avatar Sep 11 '24 16:09 GzyAftermath