Noisy output when no human present about modnet HOT 4 CLOSED

Hi, thanks for your attention.

Q1: there's always some noise in real input images
Which pre-trained models or demos have you tried? Can you show me some examples of failures?
For our image matting model, we trained it for photographic images. The pre-trained model may fail in some daily photes since we have not adapt it to the domain of daily photos.
For our video matting model, if you follow the demo guideline, I think you can get pretty good results through WebCam.

Q2: Is it possible to increase confidence score, or some post or preporcessing method to improve the results.
Yes, of course. You can set a confidence score to detect t the presence of persons in the image. I am sorry that our method can not handle images without persons, because we do not consider such a case (this is an engineering problem).

from modnet.

请问video matting model能用来做图像matting吗？怎么用呢？

from modnet.

@juzuo
你好，请将image matting demo中的：

!gdown --id 1mcr7ALciuAsHCpLnrtG_eop5-EYhbCmz \

替换为：

 !gdown --id 1Nf1ZxeJZJL8Qx9KadcYYyEmmlKhTADxX \

并重新运行代码。

from modnet.

好的，谢谢作者。

from modnet.