| 内容提要: |
The traditional Single Image Super-Resolution problem is defined as recovering a high-resolution (HR) image from its low-resolution (LR) observation. Most existing methods still suffer from blurry results at large upscaling factors(eg.x4). Perceptual-related constraints, and adversarial loss, have been introduced to the SISR problem formulation. However, they tend to hallucinate fake textures and even produce artifacts. we propose a new RefSR algorithm, named Super-Resolution by Neural Texture Transfer (SRNTT), which adaptively transfers textures from the Ref images to the SR image. |