ZhengPeng7/BiRefNet

Not getting the same quality as Hugging face demo

Closed this issue · 3 comments

Dear Author,

Thank you for your amazing work.

I tried deploying and running your code in my local with the same requirements as mentioned in the hugging face repo and using the same model that is BiRefNet-massive-epoch_240.pth but still we get a different quality .

We are using cuda : 12.3 and python 3.10.

Can you please guide me what I am doing wrong so I can get the same quality as hugging face demo?

Thanks in advance.

Hi, thanks a lot for your interest :)
You mean you got different results with the same weights (even lower quality)? That's a problem.
Could you use the single_image_inference colab to have a try and provide the images of them for me to look into the problem?
Thank you very much, looking forward to hearing from you.

Hey, how is it going? I do care about the differences of using different versions of libs.

Re-open it if you still have any trouble.