The HIDE dataset is mainly proposed for the problem of dynamic deblurring. Various blurs are caused by the relative motion between the imaging device and the scene, mainly including camera shake and object movement. In order to fully capture the dynamic blur caused by passive device interference and active behavior, the HIDE dataset came into being to cover a wider range and closer scenes, and solve the problem of motion deblurring that can be perceived by the human body.

The dataset consists of 8,422 blurry-clear image pairs and 65,784 densely annotated FG human bounding boxes.