the gap between your realization and the paper in autoslim

[priscillapan]

In the paper， AutoSlim-MobileNet-v2 at 305M FLOPs achieves 74.2% top-1 accuracy, 2.4% better than default MobileNet-v2 (301M FLOPs），but your realization is only 72.73% at 320M FLOPs。

What do you mean by "Note that we ran the official code and the Top-1 Acc of the models with official channel cfg are 73.8%, 72.5% and 71.1%"? It should be 74.2%. I test it using paper's official code and official channel cfg(https://github.com/JiahuiYu/slimmable_networks). The result is 74.2% not 72.5%

[pppppM]

This result is obtained by retraining with the official code, not the test result using the official checkpoint.

[priscillapan]

I see. Thanks. I‘ll try it later. Now I'm training subnet at 200M Flops by mmrazor :)