Training Results on CIFAR10 Dataset with an Imbalance Ratio of 100 Are Lower than the Performance Reported in the Paper

[fantastice]

Sure! Here’s the translation:

"Hello, I ran your code for 1200 epochs on the CIFAR10 dataset with an imbalance ratio of 100, but I found that the final results are lower than the data reported in the paper."

[pradipto111]

Hi @fantastice ,

Thanks for your interest in our work. This slight variation in numbers can be because of precision variations due to different GPU architecture, configurations and even CUDA versions.

You may try running different seeds to achieve the best possible performance.

Best.

[fantastice]

Thank you very much,I will try more best.