Real-World Blur Dataset for Learning and Benchmarking Deblurring Algorithms

Abstract

Numerous learning-based approaches to single image deblurring for camera and object motion blurs have recently been proposed. To generalize such approaches to real-world blurs, large datasets of real blurred images and their ground truth sharp images are essential. However, there are still no such datasets, thus all the existing approaches resort to synthetic ones, which leads to the failure of deblurring real-world images. In this work, we present a large-scale dataset of real-world blurred images and their corresponding sharp images captured in low-light environments for learning and benchmarking single image deblurring methods. To collect our dataset, we build an image acquisition system to simultaneously capture a geometrically aligned pair of blurred and sharp images, and develop a post-processing method to further align images geometrically and photometrically. We analyze the effect of our post-processing step, and the performance of existing learning-based deblurring methods. Our analysis shows that our dataset significantly improves deblurring quality for real-world low-light images.