Continuous-time method recovers sharp video from blurred images
PickMoment: Continuous-Time Single-Image-to-Video via Learning Deblurring and Blur-to-Video
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Photos taken with motion blur happen because the camera captures light over a short period, mixing multiple moments into one blurry picture. The authors created a method that learns how to undo this blur and can generate sharp images or even videos showing the moments inside the blur. Unlike previous methods that only restore one snapshot or a fixed number of frames, their approach can predict images from any point in time during the exposure. This means one model can do deblurring and make videos without needing separate training for each task.
What this means in practice
- •For camera software developers: Improve smartphone or camera firmware to recover sharp images or smooth video from a single blurred photo using a unified neural network model.$Commercial implications: Enables enhanced photo and video features for consumer devices, providing higher quality outputs without extra hardware.
- •For video editors and visual effects teams: Generate intermediate sharp frames and continuous videos from blurred footage to aid in editing or visual effect creation.
Authors
Junseong Shin, Hyeonsu Jo, Daehyun Kim, Tae Hyun Kim
Abstract
Motion blur arises from the temporal integration of a continuous sharp signal over a finite exposure window, yet existing learning-based methods sidestep this physical model and predict only the sharp signal itself: most single-image deblurring methods recover a single frame at the exposure center, while blur-to-video methods predict a fixed set of frames. We introduce PickMoment, a continuous-time reformulation that directly learns the interval-mean blur over arbitrary sub-intervals of the exposure with a single deterministic model. Drawing an analogy to MeanFlow's average-velocity formulation, we train the model with three supervisions derived from the blur integral: an empirical reconstruction loss from available subframes, an additivity loss that enforces self-consistency across overlapping sub-intervals, and a sharp-frame loss anchored at the zero-interval limit. A single trained model unifies single-image deblurring, blur-to-video generation, and continuous-time pick-a-moment recovery as different queries to the same network, with no separate training for each task. Our PickMoment achieves state-of-the-art performance among generative-based deblurring methods on GoPro and HIDE while competitive against restoration-based methods on RealBlur, and the highest per-frame fidelity on GoPro-7 blur-to-video, all in a single forward pass without iterative sampling.