Why aren't these mask images binary?

[yyywxk]

I found these mask images were not composed of 0 and 255 when I used cv2.imread to input them. So why not use binary images as mask images?

[nie-lang]

I have checked the dataset after seeing this issue. There are other pixel values (except 0 and 255) near the boundaries. Your suggestion would be more reasonable, but it would require extensive human efforts to correct them.

From my experience, it makes little difference in the performance even using binary images. Furthermore, even without the mask, the network can be trained successfully. In my view, the network predicts the initial mesh according to the input image rather than the mask.

[yyywxk]

ok, thank you!