New Smartphone Video Dataset Features 42,000 Real-World Deblurring Image Pairs
July 24, 2026
A new dataset provides 42,000 high-resolution blur-sharp image pairs derived from smartphone slow-motion videos. It is 10 times larger than existing benchmarks and includes 8 times more diverse scenes, revealing significant performance drops in current state-of-the-art deblurring models.
HOW THIS AFFECTS YOU
●
builderYou can test your computer vision pipelines against more realistic motion blur scenarios found in mobile video.
●
researcherYou can use this larger, more diverse dataset to benchmark and train more robust deblurring architectures.