szq0214 / un-mix Goto Github PK

Hello!
I wonder how the image after mixup will be used?
As positive samples or negative ones?
Thank you!

Because SWAV is not the same as MOCO, I sincerely want to know how to design the L_M Loss with SWAV

Hello, I'm very interested in your work. I want to know L_M loss， can you tell me the calculation method ?

Hi，thanks for your excellent work~
Can you provide the code combined with SimCLR?
Which conference received your paper？

Thanks for your nice work.

Could you please provide full code base for training and test on any method (Moco or BYOL or SimCLR)?

Hi, thank you very much for your work showing the good properties exhibited by the use of mix in self-supervision learning. I try to reproduce your idea in BYOL, but I have some confusions: 1. In case of self-mixtures, is it better to use mix images in momentum branches or branches with gradient updates? If you have results for 200 epochs would be much appreciated. Thanks again for your work!

I just come to show my admiration.

I find this project accidentally and I think it is very elegant and kinda under-estimated. Coincidentally, I used a similar idea in class-incremental learning and told a story, but I still like your expression more.

I really hope one day I can work with you! Maybe just hope...

Hi All, we just added a demo of UnMix in the repo for your quick trial. Please check it out!!

Also, please feel free to let me know if you have any questions or comments.