MemNet: A Persistent Memory Network for Image Restoration

• DOI: 10.48550/arxiv.1708.02209
• Publication Date: 2017-01-01
• Authors: Ying Tai; Jian Yang; Xiaoming Liu; Chunyan Xu
• Keywords: Code (set theory); JPEG

Recently, very deep convolutional neural networks (CNNs) have been attracting considerable attention in image restoration. However, as the depth grows, long-term dependency problem is rarely realized for these models, which results prior states/layers having little influence on subsequent ones. Motivated by fact that human thoughts persistency, we propose a persistent memory network (MemNet) introduces block, consisting of recursive unit and gate unit, to explicitly mine through an adaptive learning process. The learns multi-level representations current state under different receptive fields. outputs from previous blocks are concatenated sent adaptively controls how much states should be reserved, decides stored. We apply MemNet three restoration tasks, i.e., denosing, super-resolution JPEG deblocking. Comprehensive experiments demonstrate necessity its unanimous superiority all tasks over arts. Code available at https://github.com/tyshiwo/MemNet.