Reproduce “Deep Learning Empowered Super-Resolution: A Comprehensive Survey and Future Prospects” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Le Zhang, Ao Li, Qibin Hou, Ce Zhu, Yonina C. Eldar
This survey paper provides a comprehensive review of deep learning techniques for super-resolution, unifying single image, video, stereo, and light field modalities under a consistent taxonomy based on network backbones like CNNs, Transformers, GANs, and Diffusion Models. It systematically analyzes over 250 methods up to early 2025, detailing their methodologies, datasets, evaluation metrics, and empirical results, while also identifying future research directions.
Key insights
- Deep learning has largely overcome the limitations of traditional SR, enabling automatic learning of mapping functions and superior feature extraction for higher-fidelity results.
- A consistent trade-off exists between pixel-wise accuracy (often achieved by regression models like Transformers) and perceptual realism (often achieved by generative models like GANs and Diffusion Models).
- Specialized multi-modal SR approaches (VSR, SSR, LFSR) consistently leverage spatio-temporal, cross-view, or angular information to outperform generic single-image methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.