Questions about DMUE training on different datasets

[Delete12137]

Thank you very much for your work. I achieved the performance shown in the paper on the AffectNet dataset, but for the RAFDB dataset, I modified some parameters in the config.py to fit the RAFDB dataset, but the best performance was only 83%, which is far from the results in the paper. In addition to some parameters in the config.py that need to be modified, what other areas of the code need to be changed?
ps. I performed the same cropping and alignment operation as the AffectNET dataset on the original version of the RAFDB dataset (i.e. the aligned images not provided by the author) and modified the following parameters: num_classes=7, ramp_a=9/10,

[Arsiuuu]

Hi~
Have you solved it? : )

[Delete12137]

Sorry, I didn't solve it.发自我的 iPhone在 2023年4月18日，20:18，Arsenever ***@***.***> 写道： Hi~ Have you solved it? : ) —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>

[Arsiuuu]

请问您是就改了num_classes=7, ramp_a=9，以及用了aligned的rafdb，其他不变就跑出83了吗