"Our approach works by predicting a dense motion flow field and a monocular depth map directly from a single motion-blurred image"