Hello, thank you for sharing this great work.
I have a question regarding the reported performance on CIFAR-100.
According to the paper, the accuracy under the practical setting is around 46–47%. However, when I reproduced the experiment using the provided implementation, the accuracy I obtained is around 43%.
I tried to follow the settings described in the paper as closely as possible, but there is still a noticeable gap in performance.
Could you let me know if there are any important implementation details, hyperparameters, or training settings that might affect this result?
Thank you for your help.
Hello, thank you for sharing this great work.
I have a question regarding the reported performance on CIFAR-100.
According to the paper, the accuracy under the practical setting is around 46–47%. However, when I reproduced the experiment using the provided implementation, the accuracy I obtained is around 43%.
I tried to follow the settings described in the paper as closely as possible, but there is still a noticeable gap in performance.
Could you let me know if there are any important implementation details, hyperparameters, or training settings that might affect this result?
Thank you for your help.