What if I want to do imagenet transfer learning on my own data set

[tianle-BigRice]

Do I only need to download the model parameters of VIT-H_14? If this is the case, the connection you provided seems to have no model parameters.https://github.com/rwightman/pytorch-image-models

[lucidrains]

Ross framework should allow you to use ViT and download the pretrained models with one command

[tianle-BigRice]

I have already laid down the parameters of the VIT-H14 pre-trained model. I would like to ask you if it is possible to import this parameter into your definition model. ross' framework I see that I don't need to finetune it, and his framework collection has too many things in it that I don't need, so it is too cluttered.Also, are you Chinese?

[duanyaohui]

Have you ever solved it? I have a same question about how to finetune it to my own dataset with this ViT model.

[tianle-BigRice]

Have you ever solved it? I have a same question about how to finetune it to my own dataset with this ViT model.

没有，我直接换了一套demo：https://github.com/jeonsworld/ViT-pytorch，你也可以试试这个，我目前还没有跑通，但是我看了看代码应该是能用的。