Discovering Distinctive "Semantics" in Super-Resolution Networks
This work addresses fundamental questions about the working mechanisms of super-resolution networks, which is incremental but important for improving their development and generalization to real-world data.
The paper investigates whether super-resolution networks learn semantic information or just perform complex mapping, discovering that they encode deep degradation representations related to image degradation rather than content, and applies these findings to tasks like distortion identification and blind SR with promising results.
Image super-resolution (SR) is a representative low-level vision problem. Although deep SR networks have achieved extraordinary success, we are still unaware of their working mechanisms. Specifically, whether SR networks can learn semantic information, or just perform complex mapping function? What hinders SR networks from generalizing to real-world data? These questions not only raise our curiosity, but also influence SR network development. In this paper, we make the primary attempt to answer the above fundamental questions. After comprehensively analyzing the feature representations (via dimensionality reduction and visualization), we successfully discover the distinctive "semantics" in SR networks, i.e., deep degradation representations (DDR), which relate to image degradation instead of image content. We show that a well-trained deep SR network is naturally a good descriptor of degradation information. Our experiments also reveal two key factors (adversarial learning and global residual) that influence the extraction of such semantics. We further apply DDR in several interesting applications (such as distortion identification, blind SR and generalization evaluation) and achieve promising results, demonstrating the correctness and effectiveness of our findings.