I think the boundary mask is calculated wrong

[ezio1320]

hello,zhanghan:
Modnet is a exciting work in matting research, and your code is helpful a lot. Thank you !
In the process of using your code, I found that there are some problems with the calculation of the bound mask.
In trainer.py , Line 146. the mask calculated as follow: boundaries = (trimap < 0.5) + (trimap > 0.5)

I think the "boundary" means the unknown area of trimap from the paper and literally. So, boundaries = (trimap==0.5)
Which is right? Does anyone else have same question?

[ezio1320]

hello,zhanghan: Modnet is a exciting work in matting research, and your code is helpful a lot. Thank you ! In the process of using your code, I found that there are some problems with the calculation of the bound mask. In trainer.py , Line 146. the mask calculated as follow: boundaries = (trimap < 0.5) + (trimap > 0.5)

I think the "boundary" means the unknown area of trimap from the paper and literally. So, boundaries = (trimap==0.5) Which is right? Does anyone else have same question?

I tried to modify the boundaries = (trimap==0.5), the performance is bad!
Finally, I find you are right. Combine the following:
pred_boundary_detail = torch.where(boundaries, trimap, pred_detail). That 's right.The variable name ‘boundaries’ confuses me.

If boundaries = (trimap==0.5), pred_boundary_detail = torch.where(boundaries ,pred_detail, trimap).