Hi,
Thank you for making the code public.
I am running it for CIFAR100 dataset and using the provided model checkpoints for the teacher model. I have formatted the dataset as mentioned in the README.
While I am able to obtain similar performance for IPC=1 and IPC=10 as mentioned in the paper, for IPC=50, I am getting 54.66% mean accuracy with 0.25% standard deviation after repeating for 5 iterations. As per the paper, accuracy should be 62.6%.
Could you please suggest what might be causing this issue?
Thank you.
Hi,
Thank you for making the code public.
I am running it for CIFAR100 dataset and using the provided model checkpoints for the teacher model. I have formatted the dataset as mentioned in the README.
While I am able to obtain similar performance for IPC=1 and IPC=10 as mentioned in the paper, for IPC=50, I am getting 54.66% mean accuracy with 0.25% standard deviation after repeating for 5 iterations. As per the paper, accuracy should be 62.6%.
Could you please suggest what might be causing this issue?
Thank you.