Warped Diffusion: Solving Video Inverse Problems with Image Diffusion Models

TL;DR

This paper introduces a novel approach called Warped Diffusion that leverages image diffusion models in a continuous function space to improve the quality and temporal consistency of solutions to inverse video problems like inpainting and super-resolution.

Contribution

The paper proposes a new perspective of viewing video frames as continuous functions and introduces a method to ensure temporal consistency using equivariance and guidance, enabling the use of image diffusion models for video tasks.

Findings

Outperforms existing techniques in video inpainting and super-resolution.

Ensures temporal consistency through a simple post-hoc guidance method.

Successfully applies state-of-the-art diffusion models to inverse video problems.

Abstract

Using image models naively for solving inverse video problems often suffers from flickering, texture-sticking, and temporal inconsistency in generated videos. To tackle these problems, in this paper, we view frames as continuous functions in the 2D space, and videos as a sequence of continuous warping transformations between different frames. This perspective allows us to train function space diffusion models only on images and utilize them to solve temporally correlated inverse problems. The function space diffusion models need to be equivariant with respect to the underlying spatial transformations. To ensure temporal consistency, we introduce a simple post-hoc test-time guidance towards (self)-equivariant solutions. Our method allows us to deploy state-of-the-art latent diffusion models such as Stable Diffusion XL to solve video inverse problems. We demonstrate the effectiveness of our method for video inpainting and $8\times$ video super-resolution, outperforming existing techniques based on noise transformations. We provide generated video results: https://giannisdaras.github.io/warped_diffusion.github.io/.