About FLOPs

[AFeng-x]

Good work！And I wanna know how you calculate FLOPs of the model in your paper? Thanks!!!

[OliverRensu]

Hi, please refer mmcv flops count https://github.com/open-mmlab/mmcv/blob/master/mmcv/cnn/utils/flops_counter.py AFeng ***@***.***> 于2022年4月24日周日 17:09写道：

…

Good work！And I wanna know how you calculate FLOPs of the model in your paper? Thanks!!! — Reply to this email directly, view it on GitHub <#5>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ANP7CCFHPL52SIGJ4XE2WJTVGUFVPANCNFSM5UF6YBBA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Andy1621]

I remember that the mmcv FLOPs count does not support attention calculation, which will lead to lower FLOPs. I suggest tot use fvcore as in our UniFormer.

[OliverRensu]

Yes, we need to manually calculate the flops of Attention but it is not very large with the input of 224x224. Kunchang Li ***@***.***> 于2022年5月28日周六 02:35写道：

…

I remember that the mmcv FLOPs count does not support attention calculation, which will lead to lower FLOPs. I suggest tot use fvcore as in our UniFormer <https://github.com/Sense-X/UniFormer/blob/ac62a49b40b79e614501905ad2f1c19bd4dcf12e/image_classification/main.py#L258-L259> . — Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ANP7CCCR6VQRQJWDPEGLPI3VMEIVLANCNFSM5UF6YBBA> . You are receiving this because you commented.Message ID: ***@***.***>