Zhiqiang Tang

I am also wondering the same question here. It seems you just do the normal BN initialization and usage. There is no code related to how separated mean and var...

@janeyoung2018 You can try our new AutoMMPredictor (already merged into the main branch) for image+tabular regression. Basically, AutoMM can cover any combinations of image, text, numerical, and categorical features. For...

Please make sure your Pytorch version matches the corresponding script. Script names with "prev-version" require Pytorch v0.1.12.

Hi @weix-liu, AdaIN was used to generate the Stylized ImageNet, which is a comparison method in our paper. We also analyzed the difference between CrossNorm and Stylized ImageNet in the...

Hi @Keysmis, The adversarial data augmentation is a general method. It can improve the performance of different human pose estimators. I think you can replace the hourglass networks in the...

Why needs to make timm optioinal? It's a core dependency of AutoMM.