The details of the pre-training.

[Swecamellia]

I am very interested in the pre-training process. Could you please provide the relevant code details?
That way, I can apply it to my own dataset. Thank you very much!

[BodongDu]

I have the same request with you, I am looking forward to your reply!

[bighan123]

based on the paper details (Fig 2), I think the pre-training way is autoregressive modeling (like MAE), and the main difference is that the author try to predcit token both in spatial and frequency domains, but i am not sure wherer the code is entirely based on the paper description.