Revision history [back]

I am trying to understand if the gradient and connectivity errors for natural image matting presented in the paper:

"A Perceptually Motivated Online Benchmark for Image Matting"

have to be divided by the number of pixels in the trimap region.