Deep edge map guided depth super resolution

作者：

Highlights：

• Edge subnetwork using RGB-D information infers edge maps to assist super-resolution.

• SR subnetwork proposes a weight-sharing module to extract the general features.

• SR subnetwork proposes an adaptive module to adapt to the specific features.

• We construct a benchmark dataset to facilitate recovery of real-world depth images.

摘要

•Edge subnetwork using RGB-D information infers edge maps to assist super-resolution.•SR subnetwork proposes a weight-sharing module to extract the general features.•SR subnetwork proposes an adaptive module to adapt to the specific features.•We construct a benchmark dataset to facilitate recovery of real-world depth images.

论文关键词：Super resolution,Depth map,Edge prediction,Disentangling

论文评审过程：Received 5 March 2020, Revised 5 September 2020, Accepted 13 October 2020, Available online 20 October 2020, Version of Record 1 November 2020.

论文官网地址：https://doi.org/10.1016/j.image.2020.116040