DA-Flow: Degradation-Aware Optical Flow Estimation with Diffusion Models

Abstract

Optical flow models trained on high-quality data often degrade severely when confronted with real-world corruptions such as blur, noise, and compression artifacts. To overcome this limitation, we formulate Degradation-Aware Optical Flow, a new task targeting accurate dense correspondence estimation from real-world corrupted videos. Our key insight is that the intermediate representations of image restoration diffusion models are inherently corruption-aware but lack temporal awareness. To address this limitation, we lift the model to attend across adjacent frames via full spatio-temporal attention, and empirically demonstrate that the resulting features exhibit zero-shot correspondence capabilities. Based on this finding, we present DA-Flow, a hybrid architecture that fuses these diffusion features with convolutional features within an iterative refinement framework. DA-Flow substantially outperforms existing optical flow methods under severe degradation across multiple benchmarks.

Figures (13)

• Figure 1: Qualitative comparison of DA-Flow with baselines teed2020raftwang2024seapoggi2025flowseek on Spring benchmark Mehl2023_Spring. Under severe degradations, existing optical flow methods fail to recover reliable correspondences, whereas our DA-Flow accurately estimates the underlying motion.
• Figure 2: Overall architecture of DA-Flow.
• Figure 3: Comparison of zero-shot geometric correspondence between Baseline and Lifting features. (a) Top-10 layers ranked by timestep-averaged EPE (lower is better). Lifting consistently achieves lower EPE across all ranks. (b) EPE over denoising steps for the top-4 layers of each method. Baseline features show high sensitivity to the denoising step, while Lifting features remain stable across the denoising steps.
• Figure 4: Qualitative results on Sintel butler2012naturalistic.
• Figure 5: Qualitative results on Spring Mehl2023_Spring.