Is the segmented result right?

[tengshaofeng]

@lizhengwei1992 , Thanks so much for your contribution. u really did a great job.
when I run the test_camera.sh in branch of semantic_human_matting-fast_human_matting, the result seems not ideal.
my input image is :

and the output frame_seg is:

note: I have commet the cv2.capture, instead, I add cv2.imread the input image above