Diffusion Models, Image Super-Resolution And Everything: A Survey

Diffusion Models (DMs) have disrupted the image Super-Resolution (SR) fieldand further closed the gap between image quality and human perceptualpreferences. They are easy to train and can produce very high-quality samplesthat exceed the realism of those produced by previous generative methods.Despite their promising results, they also come with new challenges that needfurther research: high computational demands, comparability, lack ofexplainability, color shifts, and more. Unfortunately, entry into this field isoverwhelming because of the abundance of publications. To address this, weprovide a unified recount of the theoretical foundations underlying DMs appliedto image SR and offer a detailed analysis that underscores the uniquecharacteristics and methodologies within this domain, distinct from broaderexisting reviews in the field. This survey articulates a cohesive understandingof DM principles and explores current research avenues, including alternativeinput domains, conditioning techniques, guidance mechanisms, corruption spaces,and zero-shot learning approaches. By offering a detailed examination of theevolution and current trends in image SR through the lens of DMs, this surveysheds light on the existing challenges and charts potential future directions,aiming to inspire further innovation in this rapidly advancing area.

Further reading#

• Access Paper in arXiv.org