DRealSR, the full name of Diverse Real-world image Super-Resolution, is a super-resolution image dataset about the degradation process of real images, alleviating the limitations of traditional simulated image degradation.

The images were collected by a digital SLR camera in 5 natural scenes, covering indoor and outdoor scenes such as advertising posters, plants, offices and buildings, avoiding moving objects. The images of the training set were cropped into image patches of 380×380, 272×272 and 192×192 sizes, for a total of 31,970 image patches. The test set has 93 image pairs.