Multi-scale foreground-background separation for light field depth estimation with deep convolutional networks

Jae Young Lee, Jiwan Hur, Jaehyun Choi, Rae-Hong Park, Junmo Kim

Abstract: Highlights•For depth from light field, a framework based on the foreground-background separation with deep learning is presented.•A method of the multi-scale foreground-background separation is proposed.•To further improve around occlusion boundary, a loss function matching gradient magnitude is proposed.