Settings for fine-tuning on CUB dataset

[higersky]

I'm trying to reproduce your fine-tuning results on MAE. What are your training settings for fine-tuning on the CUB dataset?
I used the official pre-trained model of MAE-Base and followed the default settings on ImageNet-10K. I got 84.19 in acc-1. So I am very curious about your settings to get 86.5.
Thank you!

[ucasligang]

We use 'vit_base_patch8' as the backbone model instead of 'vit_base_patch16' for the task of fine-grained image recognition.