HIDE

A large-scale dataset for human-aware motion deblurring, consisting of 8,422 pairs of blurry and sharp images with 65,784 densely annotated FG human bounding boxes.

BibTex: