Improvement Suggestions for Hair Matting in RMBG-2.0
Hello Team,
I hope this message finds you well. I wanted to start by conveying my admiration for the exceptional work on the RMBG-2.0 model. Its capabilities in general object matting are truly remarkable.
However, I've noticed that the hair matting performance seems to lag behind some of the previous methods, as demonstrated in this insightful blog post:
(https://hadrien-montanelli.github.io/2022-12-30.html)
Is the perceived decrease in hair matting quality an inevitable consequence of the general background matting methodology as compared to more specialized human matting techniques? If so, I'm interested in understanding if there are any strategies in place or upcoming enhancements planned that could potentially improve the hair matting performance specifically within the RMBG framework.
For your reference, here are the matting results produced by RMBG-2.0:
The matting results by RMBG-2.0:
I believe that by addressing this specific area, RMBG-2.0 could achieve even greater accuracy and detail in image matting tasks.
Looking forward to your thoughts on this matter.
Best regards,