shashankvkt / alignmixup_cvpr22 Goto Github PK

hello! It's an excellent work, and could you provide the code about training the auto-encoder please? Thank you very much!

Also, if you do not use AE, you do not use joint training as claimed in the paper, am I correct on this？

Hello. Thank you for your great work.

Currently, I'm trying to reproduce vanilla performance with your imagenet_fast protocol. However, my model 2-3% behind with your configuration. Can you share tricks (data augmentation or optimized configuration) or training log?

Thanks for your contributions to the literature.

Does the code include only mixup? I cannot find the encoder part unfortunately.

Hi, @shashankvkt. Thanks for your great work in CVPR'2022. We have supported your AlignMix in OpenMixup (issue 35) and we will train AlignMix for our mixup benchmark on CIFAR-10/100, Tiny-ImageNet, and ImageNet in the next few weeks. We will update the results in Model_Zoo_sup when we finish training. Please feel free to contact me, and you can close this issue if you have no questions about our implementations.

Have a nice day😊
Siyuan Li

Hi, thank you so much for this nice work, would you like to provide the training logs for ImageNet ?

Hi authors,
Thanks for your great work. I have noticed AlignMix since last year and have some questions about image classification results on ImageNet-1k. I notice that AlignMix reports top-1 accuracy of 81.17 on ImageNet-1k based on ResNet-50 training 300 epochs, but this repo reports 79.32. Why? I am looking forward to the open-source and new pre-print.