DiTVR: Zero-Shot Diffusion Transformer for Video Restoration

2025-08-11

AI Summary

Video restoration faces challenges including strong reliance on paired training data, distorted detail generation, and temporal inconsistency. To address these, we propose an unpaired diffusion Transformer framework. Our key contributions are: (1) a trajectory-aware attention mechanism that aligns tokens along optical flow paths to enforce motion consistency; (2) a wavelet-guided sampling strategy with optical flow regularization, integrating low-frequency priors and motion cues to improve zero-shot reconstruction quality; and (3) dynamic spatiotemporal neighborhood caching coupled with sensitivity-aware layer prioritization, enhancing temporal dynamics modeling. The method significantly improves detail fidelity and inter-frame consistency across video super-resolution, denoising, and deblurring tasks. It demonstrates robustness to optical flow noise and occlusions, achieving state-of-the-art zero-shot performance on multiple benchmarks.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Multimodal Learning
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Video restoration aims to reconstruct high quality video sequences from low quality inputs, addressing tasks such as super resolution, denoising, and deblurring. Traditional regression based methods often produce unrealistic details and require extensive paired datasets, while recent generative diffusion models face challenges in ensuring temporal consistency. We introduce DiTVR, a zero shot video restoration framework that couples a diffusion transformer with trajectory aware attention and a wavelet guided, flow consistent sampler. Unlike prior 3D convolutional or frame wise diffusion approaches, our attention mechanism aligns tokens along optical flow trajectories, with particular emphasis on vital layers that exhibit the highest sensitivity to temporal dynamics. A spatiotemporal neighbour cache dynamically selects relevant tokens based on motion correspondences across frames. The flow guided sampler injects data consistency only into low-frequency bands, preserving high frequency priors while accelerating convergence. DiTVR establishes a new zero shot state of the art on video restoration benchmarks, demonstrating superior temporal consistency and detail preservation while remaining robust to flow noise and occlusions.

Problem

• Ensuring temporal consistency in video restoration tasks
• Overcoming limitations of traditional regression-based methods
• Addressing challenges in generative diffusion models for videos

Innovation

• Diffusion transformer with trajectory aware attention
• Wavelet guided flow consistent sampler
• Spatiotemporal neighbor cache for token selection