This work proposes a method that bridges the gap between the performance of trimap-based networks and the convenience of trimap-free networks. Here image matting is divided into three subproblems: 1) image segmentation, 2) trimap generation, and 3) trimap-based image matting. The result is a trimap-free method for image matting with the added advantage that user can guide the matting procedure by tweaking the generated trimaps. Our assumption is that “fixing” a trimap is far less work than “fixing” the matte post hoc.
A modular design
Comparison with P3M-Net and ModNet schemes
Comparison with P3M-Net scheme
Left to right: image, P3M-Net, ours, ground truth.
For technical details please look at the following publication(s)