May be a small bug...

[Jiaozrr]

I tried several times to train the M_net, but it always failed to generate corect alpha. I find a possible mistake in the 36th line of pytorch_semantic_human_matting/model/network.py. It is written that bg, fg, unsure = torch.split(trimap, 1, dim=1). However, after generating some images to show the bg, fg and unsure, I found fg and unsure seemed to be reversed. I changed this line of code into: bg, unsure, fg = torch.split(trimap, 1, dim=1) and fixed my problem.

[CabbageWust]

@Jiaozrr Hi ， can you share the datasets with me ?

[Jiaozrr]

@Jiaozrr Hi ， can you share the datasets with me ? Hi, I use the dataset publicly posted at https://github.com/aisegmentcn/matting_human_datasets, mentioned in another implementation by lizhengwei1992.

[CabbageWust]

@Jiaozrr Thx

[tobechao]

@Jiaozrr Hi, What is the effect of your training t-net， I use the aifenge data training effect is very poor？

[Jiaozrr]

@Jiaozrr Hi, What is the effect of your training t-net， I use the aifenge data training effect is very poor？

Hi, I also get a poor t-net. I guess this dataset is not good enough.

[sanshibayuan]

I think you're right. But I found the author's comment in another SHM repo https://github.com/lizhengwei1992/Semantic_Human_Matting/issues/10

I'm not sure what he means……