Lightweight Degradation-Aware Super-Resolution (LDASR) Framework

We introduce a custom high-resolution image dataset designed for training and evaluating image super-resolution and restoration models under diverse real-world degradations. The dataset is constructed by combining and carefully filtering images from established sources including DIV2K, Flicker2K, and OST. To maintain high-quality visual information, we retained only images with a resolution above 480\u00d7480 pixels and a file size greater than 500 KB, removing low-quality or redundant samples. Our final collection provides a clean, diverse, and high-fidelity set of real-world images suitable for robust model training. The corresponding low-resolution counterparts are synthesized using a multi-order degradation pipeline, enabling flexible and realistic degradation scenarios. For standardized evaluation, we include results and compatibility with popular benchmarks. This dataset aims to support research in high-quality image restoration and super-resolution by offering a reliable and versatile foundation for learning-based methods.