Closed JWSunny closed 1 month ago
Since AEMatter's paper has not been officially published, there are still some issues to be resolved. We will not be releasing the training code for now.
It should be noted that we do not give the droppath probabilities for training in the model. In addition, we follow FBAmatting to generate trimap which is different from MatteFormer.
OK,thanks
I have started to compile the training code, which is expected to be made public in June.
The training codes are now available.
thanks for your sharing work, can you release the train code,I integrated the model structure into Matteformer, referred to the experimental parameters in AEMatter's paper, and retrained AEMatter. However, I found that the testing was not ideal.