How to use the pretrain model

[ankan8145]

Can anyone tell me how to use the checkpoint of "ViT-Base/16" model ? and what is mask ratio here mask_ratio=1 means full mask the full or mask_ratio=0 means full mask the full image ?

[SaumyaBhandari]

mask_ratio=1 means full mask, as the paper mentions that they have used 95% masking, which is termed as 0.95 mask_ratio.

You can look at my closed issue here where I've given the code to seamlessly load the weights from the pretrained model.

Just include the filepath of the checkpoint in args.finetune.