From general to specific: Online updating for blind super-resolution

Most deep learning-based super-resolution (SR) methods are not image-specific: 1) They trained on samples synthesized by predefined degradations (e.g. bicubic downsampling), regardless of the domain gap between training and testing data. 2) During testing, they super-resolve all images same set model weights, ignoring degradation variety. As a result, most previous may suffer performance drop when test unknown various (i.e. case blind SR). To address these issues, we propose an online SR (ONSR) method. It does rely allows weights to be updated according image. Specifically, ONSR consists two branches, namely internal branch (IB) external (EB). IB could learn specific given LR image, EB super resolve degraded learned degradation. In this way, customize for each thus get more robust degradations. Extensive experiments both real-world show that can generate visually favorable results achieve state-of-the-art in SR.