Input size in pre-trained and paper

[EoinKenny]

Hi, thanks a lot for the excellent repo!

A small question. The input size on the pre-trained models ( https://github.com/Alibaba-MIIL/ImageNet21K/blob/main/MODEL_ZOO.md) is 224x224, but your paper seems to use 384x384?

Am I missing anything? Thank you

[chou141253]

We use swin-transformer pretrained model which named 'swin_large_patch4_window12_384_in22k'. And Vit model using 'vit_base_patch16_224_miil_in21k' as pretrained model, which use 224x224 as input size, but we expand position encoding parameters to fit 448x448.