Enrico Fini

I agree this makes a lot of sense. We will try to implement it soon. It is quite easy to do, you just need to pass a dictionary with two...

We should keep in mind that [this repo](https://github.com/facebookresearch/suncet/blob/main/src/data_manager.py) has nice parameters for multi-crop on CIFAR Edit: I have just tested their parameters and they work quite well

Hi, 1) yes it was a typo, I have just fixed it. 2) Ok, can you provide the script you used and the full output error? Offline linear evaluation does...

Hi @yuweihao, thanks for the nice reply. Apex can be hard to install without sudo, that is why I prefer native AMP. Actually, I have tried both (Apex, native) with...

Thank you again! Looking forward to the release!

Hey @yuweihao, sorry to bother you again. For the hybrid stage [Pool, Pool, Attention, Attention] did you use layer norm just for the attention blocks or for the pooling blocks...

For anyone wondering, I got the following results on ImageNet-100: * vanilla PoolFormer (`return self.pool(x) - x`): 87.64 * simple pooling (`return self.pool(x)`): 87.56 * DWConv (`return dwconv(x)`): 88.10

Hi, I have just checked my logs and the results seem consistent with the ones we published. See the screenshot below: The second run (65.4%) is with slightly different hyperparams....

Yes, it is the accuracy of the final model as reported in the paper. Intermediate checkpoints are only used for forgetting. Please see this screenshot from the paper below: In...

I found some checkpoints that might be relevant: https://drive.google.com/drive/folders/1gOejzl4Q0cqAcmEjUhyStYPDbXPn1o9R?usp=share_link You can find the pre-train args there as well. I am not 100% sure that this is the correct checkpoint, so...