JizhiziLi / GFM

[IJCV 2022] Bridging Composite and Real: Towards End-to-end Deep Image Matting

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Boundary problem

dreamer121121 opened this issue · comments

I find that if i use the predict alpha multified with your initial image to get the color foreground, which is also the way used in your inference code, the final color foreground image will be surrounded by gray lines in boundary regions

Could you please share more details or maybe some examples of this case(in this issue or directly to my email jili8515@uni.sydney.edu.au)? Please make sure the predicted alpha does not have noise in boundary regions before generating the color foreground.

@JizhiziLi hi, do you use any foreground estimation algorithm to estimate the color foreground? It seems that you directly use the original image as the foreground image.