Reblur2Deblur: Deblurring Videos via Self-Supervised Learning

TL;DR

Reblur2Deblur introduces a self-supervised approach that fine-tunes existing video deblurring neural networks by ensuring their outputs, when re-blurred, match the original blurry input, leading to improved deblurring performance.

Contribution

It bridges physics-based and learning-based deblurring methods by fine-tuning neural networks with a self-supervised re-blurring consistency constraint.

Findings

Significant performance improvements on multiple datasets.

Enhanced visual quality and image metrics.

Effective integration of physics-based and data-driven approaches.

Abstract

Motion blur is a fundamental problem in computer vision as it impacts image quality and hinders inference. Traditional deblurring algorithms leverage the physics of the image formation model and use hand-crafted priors: they usually produce results that better reflect the underlying scene, but present artifacts. Recent learning-based methods implicitly extract the distribution of natural images directly from the data and use it to synthesize plausible images. Their results are impressive, but they are not always faithful to the content of the latent image. We present an approach that bridges the two. Our method fine-tunes existing deblurring neural networks in a self-supervised fashion by enforcing that the output, when blurred based on the optical flow between subsequent frames, matches the input blurry image. We show that our method significantly improves the performance of existing methods on several datasets both visually and in terms of image quality metrics. The supplementary material is https://goo.gl/nYPjEQ