Hold on to your lab coats, fellow scientists! This work will take you on a wild ride through the world of Super-Resolution (SR). In SR, the quest is to scale up a Low-Resolution (LR) image to a High-Resolution (HR) image. Our work will send you into a journey with the ins and outs of SR, diving into the basic concepts, state-of-the-art methods and new ideas for SR.
The work commences with an introduction to the basic definitions and terminology. It then proceeds to discuss evaluation metrics for SR solutions. Along the way, you'll encounter various datasets that provide diverse data types, like 8K resolution images or video sequences. The adventure continues as you delve into learning objectives of Super-Resolution: Regression-based SR (also with uncertainty), Generative SR, and, more recently, Denoising Diffusion Probabilistic Models.
This work dives also into the world of various upsampling methods, a critical aspect in image SR. It covers the whys and hows of interpolation-based and learning-based upsampling techniques and also problems appearing with upsampling techniques (e.g., artifacts) in supplemental material. Also, our work explores the challenges and potential solutions of flexible upsampling for real-world scenarios with arbitrary scaling factors.
You'll also learn about additional learning strategies such as curriculum learning, enhanced predictions, learned degradation, network fusion, multi-task learning, and normalization techniques like the Adaptive Deviation Modulator (AdaDM).
But wait, there's more! Gather around, as we embark on an epic quest to unveil the mystic secrets for SR models and architectures! In this treacherous journey, we will first traverse the hallowed halls of upsampling location, unearthing four known and distinct artifacts: Pre-upsampling, Post-upsampling, Progressive Upsampling, and Iterative Up-and-Down Upsampling. These relics, as displayed in a mystical illustration, shall guide our understanding of the SR models' construction.
In this wondrous land, we'll encounter these primary categories: Simple Networks, Residual Networks, Recurrent-Based Networks, Lightweight and Wavelet-based Models. We shall explore various state-of-the-art architectures and discuss their pros and cons. Finally, you will get a scoop of Unsupervised SR and Neural Architecture Search for SR! At last, but not least, we will discuss the future of SR, and the challenges that lie ahead. So, gather your party and embark on this thrilling adventure, delving into the diverse and intricate world of deep learning-based super-resolution networks and unlock its secrets! Spoiler-Warning: Big tables included.
