Waterloo Exploration Large-Scale Image Quality Assessment Database

*This dataset supports online use.Click here to jump.

The Waterloo Exploration Database is a large-scale image quality assessment (IQA) database created by the University of Waterloo in Canada. It aims to provide challenges to the diversity of real-world digital image content and test the generalization ability of image quality assessment models. The database contains 4,744 original natural images and 94,880 distorted images created from these original images.

Unlike the method of collecting the average opinion score of each image through subjective testing, Waterloo Exploration Database proposes three alternative test criteria to evaluate the performance of IQA models, namely the original/distorted image discernibility test (D-test), the list sorting consistency test (L-test), and the pairwise preference consistency test (P-test). Using these criteria, researchers compared 20 well-known IQA models, and the results showed that even for the best-performing no-reference IQA model, more than 6 million model failure cases can be automatically found in more than 1 billion test pairs.