How to Fine-tune MS2L ?

Question

How to Fine-tune MS2L ?

yanghhx opened this issue 3 years ago · 2 comments

Your work is very great and I am very interested in it.

I'd like to know how to fine tune MS2L for evaluation? Have the fine-tuning step is included in your code? What is the differrence between the fine-tuning protocol and the linear protocol?

Answer 1 · 2022-04-10T07:17:22.000Z

Thank you for your attention! Fine-tuning updates all network weights, including the encoder and classifier. Linear evaluation, on the other hand, only updates the classifier, and its encoder weights are fixed.

Answer 2 · 2022-04-11T02:33:48.000Z

Thanks for your replication. I will have a try. At 2022-04-10 15:17:32, "LanglandsLin" ***@***.***> wrote: Hi @yanghhx ! Thank you for your attention! Fine-tuning updates all network weights, including the encoder and classifier. Linear evaluation, on the other hand, only updates the classifier, and its encoder weights are fixed. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>