Deep Bilateral Learning for Stereo Image Super-Resolution

Bilateral filter has demonstrated its effectiveness in many traditional methods for image restoration tasks. In this letter, we incorporate the idea of bilateral grid processing a CNN framework and propose stereo super-resolution network (BSSRnet). Specifically, use parallax-attention module to information from left right views learn content-aware filters. Then, these filters are used recover missing details at different spatial locations while preserving consistency. Our is fully differentiable robust both content disparity variations. Comparative results show that our BSSRnet achieves state-of-the-art performance on Flickr1024, Middlebury, KITTI 2012 2015 datasets. Source code available at.