SATB-VR: Training Few-Step Video Restoration Diffusion Model using SNR-Aware Trajectory Blending

2026-06-26

AI Summary

Existing few-step video restoration diffusion models struggle to reconstruct fine textures when optimizing for efficiency and suffer from a mismatch between training and inference. To address these issues, this work proposes an efficient few-step video restoration method that employs an auxiliary predictor to skip initial denoising steps with low signal-to-noise ratio (SNR) and introduces an SNR-aware trajectory fusion strategy to harmonize the joint training of the predictor and the denoiser. Additionally, a denoiser-driven consistency loss is incorporated to dynamically enhance the predictor’s accuracy. The proposed approach consistently outperforms state-of-the-art methods across synthetic, real-world, and AIGC-generated video datasets under inference budgets of five steps or fewer.

Abstract

While diffusion models excel in video restoration, their reliance on extensive iterative steps limits efficiency. Conversely, aggressive single-step distillation often compromises fine texture recovery. To achieve an optimal balance, we present SATB-VR, a few-step paradigm that jump-starts the denoising process via an auxiliary predictor, explicitly bypassing early low signal-to-noise ratio (SNR) steps. However, naive joint training of the predictor and the denoiser inherently introduces a severe train-inference discrepancy. To resolve this, we propose the SNR-Aware Trajectory Blending (SATB) strategy. During the forward process, SATB constructs the noisy input by dynamically blending the predictor's output with the ground-truth trajectory based on the SNRs. This forces the denoiser to robustly compensate for initial prediction errors while smoothly converging to the clean data manifold. Furthermore, we introduce a Denoiser-Driven Consistency (DDC) loss, leveraging the concurrently updated denoiser as a dynamic evaluator to explicitly align internal features and boost predictor accuracy. Extensive experiments demonstrate that, under flexible few-step inference regimes (\eg, $\le 5$ steps), SATB-VR performs favorably against existing approaches on synthetic, real-world, and AIGC benchmarks.

Problem

• video restoration
• diffusion model
• few-step inference
• SNR
• texture recovery

Innovation

• few-step diffusion
• SNR-aware trajectory blending
• video restoration
• denoiser-driven consistency
• train-inference alignment