CasSR: Activating Image Power for Real-World Image Super-Resolution

TL;DR

A cascaded controllable diffusion model that aims to optimize the extraction of information from low-resolution images and a multi-attention mechanism to enhance the T2I model's capability in maximizing the restoration of the original image content are proposed.

Abstract

The objective of image super-resolution is to generate clean and high-resolution images from degraded versions. Recent advancements in diffusion modeling have led to the emergence of various image super-resolution techniques that leverage pretrained text-to-image (T2I) models. Nevertheless, due to the prevalent severe degradation in low-resolution images and the inherent characteristics of diffusion models, achieving high-fidelity image restoration remains challenging. Existing methods often exhibit issues including semantic loss, artifacts, and the introduction of spurious content not present in the original image. To tackle this challenge, we propose Cascaded diffusion for Super-Resolution, CasSR , a novel method designed to produce highly detailed and realistic images. In particular, we develop a cascaded controllable diffusion model that aims to optimize the extraction of information from low-resolution images. This model generates a preliminary reference image to facilitate initial information extraction and degradation mitigation. Furthermore, we propose a multi-attention mechanism to enhance the T2I model's capability in maximizing the restoration of the original image content. Through a comprehensive blend of qualitative and quantitative analyses, we substantiate the efficacy and superiority of our approach.

Figures (6)

• Figure 1: Our model has demonstrated strong image super-resolution capabilities. As shown above, CasSR has achieved outstanding results on both real-world and synthetic data.
• Figure 2: Comparison between our method and other state-of-the-art methods. Please pay attention to the highlighted areas. Our method accurately restored detailed textures of the objects. Zoom in for better view.
• Figure 3: Architecture of the proposed Cascaded Super Resolution(CasSR) network.
• Figure 4: Qualitative comparisons of different methods on real image super-resolution. Please focus on the areas marked by red boxes, and zoom in for a better view.
• Figure 5: Qualitative comparison on LQ image. CasSR successfully recovers the penguin's eye and also avoids confusing its wings with background rocks. Comparing to other methods, CasSR output preserves better details existing in the LQ image.