I recently came across the paper, “ResNet Strikes Back: An Improved Training Procedure in Timm”. This paper describes the improved methodology for training ResNets, which notably includes the use of specific data augmentations.
My question is whether this training procedure, especially the augmentation techniques, is also employed for other models within the Timm library. The models I’m particularly interested in are resnet*d, efficientnet_b*, convnext_*, vit_*, swin_*, and mobilenetv3_small_*.
Could anyone please direct me to relevant publications, or sections within the Timm codebase, which detail the specific augmentations employed in training these aforementioned models?
Thank you in advance for your assistance!