search

QLYoo / LFPNet

GNU General Public License v3.0
41 stars 10 forks source link

Retrieve foreground prediction #5

Closed WANGEOGEO closed 3 years ago

WANGEOGEO commented 3 years ago

Hi, This work looks fantastic and performs well. Recently I have been studying on foreground prediction during matting, but I am not quite sure which part of the code in eval.py is about foreground prediction? I would like to visualize those predictions for further study. Thanks in advance for help!

Windaway commented 3 years ago

There is no foreground prediction code in eval.py. You may use the foreground output of the LFPnet defined in model.py. But the predicted foreground is not good. I think GCAMatting, which uses the module from the inpainting network, may be more suitable for you.

Windaway commented 3 years ago

There is a difference between LFPNet and FBAMatting results. LFPNet only estimates the foreground in the unknown regions and the foreground regions, FBAmatting estimates foreground on the entire image. I think this is why FBAMatting can predict foreground better.

RGB: image image FBA: image image LFPNet: image image

WANGEOGEO commented 3 years ago

Ah I see ...... Great appreciation on showing some visualizations, now I have a better understanding on the mechanisms.