-
Notifications
You must be signed in to change notification settings - Fork 49
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Question regarding smoothing of edges #9
Comments
Thanks for posting this question. I think ISNET returns the high-resolution binary mask while MAM returns the continuous alpha channel values between 0 to 1, which may makes the boundary looks not as smoothing as the binary mask. One solution could be adding binarization in the end to MAM, but if you want binary mask as output I guess ISNet may fits your needs better than MAM. |
One more thing is that upsampling could return the non-smoothing boundary as well. We resize the image to 1024x1024 to MAM and upsample the final output to original input size. If you use high-res inputs like 2K/4K, which ISNET is designed for, could result in the non-smoothing boundary for MAM's output. |
HI @chrisjuniorli thanks for the reply:
My inputs are less than 1024 x 1024.
Can you elaborate on how to add the binarization to the end of MAM. My understanding is that , say get the final pixel values and make everything less than 0.5 as 0 and greater than 0.5 as 1 is extremely naive and might not be great at giving results. So would appreciate if there is anyway to include better ways to incorporate binarization in MAM. |
Hi, I am currently using IS-NET (https://github.com/xuebinqin/DIS) in my workflow. I want to use Matt Anything, however I am facing an issue. The main one being that the edges of the masking are not as smooth as the ones from IS-Net.
Input Image:
ISNET Mask:
Mask from MAM HF space:
I wanted to know how to get the IS-net mask smoothness using MAM. Please assist.
The text was updated successfully, but these errors were encountered: