Implementation of PRM module
Opened this issue · 2 comments
Hello, your work MGMatting is very excellent. I read your paper and code, I found that the PRM module code in the paper is implemented using two-layer convolution. Do you have any code for PRM in the paper? Does PRM bring a great gain effect when you make experiments, especially for the details such as the hairline of the portrait matting? Thank you very much for answering this question.
Hi, please refer to the decoder file in the code-base, where additional convs of PRM are defined. But the usage of PRM also include some code parts in utils.py and main.py, in terms of how to use later-stage prediction to refine the previous stage's one, and also related loss computation.
Hi, please refer to the decoder file in the code-base, where additional convs of PRM are defined. But the usage of PRM also include some code parts in utils.py and main.py, in terms of how to use later-stage prediction to refine the previous stage's one, and also related loss computation.
Thank you very much.In addition, I would like to ask if you encountered a situation where the background pixel was incorrectly predicted as the foreground pixel during the test (the value of the background pixel is predicted to be a low confidence value), do you know how to solve it or how to strengthen the guidance of mask for the final alpha result.