Residual Dense Network for Image Restoration

Recently, deep convolutional neural network (CNN) has achieved great success for image restoration (IR) and provided hierarchical features at the same time. However, most CNN based IR models do not make full use of from original low-quality images; thereby, resulting in relatively-low performance. In this work, we propose a novel efficient residual dense (RDN) to address problem IR, by making better tradeoff between efficiency effectiveness exploiting all layers. Specifically, block (RDB) extract abundant local via densely connected RDB further allows direct connections state preceding layers current RDB, leading contiguous memory mechanism. To adaptively learn more effective stabilize training wider network, proposed feature fusion RDB. After fully obtaining features, global jointly holistic way. We demonstrate RDN with several representative applications, single super-resolution, Gaussian denoising, compression artifact reduction, deblurring. Experiments on benchmark real-world datasets show that our achieves favorable performance against state-of-the-art methods each task quantitatively visually.