Consistency Trajectory Matching for One-Step Generative Super-Resolution

2025-03-26

AI Summary

Existing diffusion-based super-resolution methods achieve high fidelity but suffer from substantial inference overhead; mainstream knowledge distillation approaches incur prohibitive training costs and are inherently constrained by teacher model performance. Method: We propose the first distillation-free, one-step generative super-resolution framework. It establishes a deterministic probability flow ordinary differential equation (PF-ODE) mapping from low-resolution (LR) noisy images to high-resolution (HR) outputs and learns this mapping directly via consistency training. Crucially, we introduce distribution trajectory matching (DTM) loss, which enforces alignment between the HR reconstruction trajectory and the natural image PF-ODE trajectory—eliminating reliance on any teacher model. Contribution/Results: On both synthetic and real-world benchmarks, our method achieves visual quality comparable to or exceeding multi-step diffusion models in a single generation step. It significantly reduces both training and inference overhead, enabling end-to-end, distillation-free, high-fidelity super-resolution.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Distributed Machine Learning & Federated Learning
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• 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

Current diffusion-based super-resolution (SR) approaches achieve commendable performance at the cost of high inference overhead. Therefore, distillation techniques are utilized to accelerate the multi-step teacher model into one-step student model. Nevertheless, these methods significantly raise training costs and constrain the performance of the student model by the teacher model. To overcome these tough challenges, we propose Consistency Trajectory Matching for Super-Resolution (CTMSR), a distillation-free strategy that is able to generate photo-realistic SR results in one step. Concretely, we first formulate a Probability Flow Ordinary Differential Equation (PF-ODE) trajectory to establish a deterministic mapping from low-resolution (LR) images with noise to high-resolution (HR) images. Then we apply the Consistency Training (CT) strategy to directly learn the mapping in one step, eliminating the necessity of pre-trained diffusion model. To further enhance the performance and better leverage the ground-truth during the training process, we aim to align the distribution of SR results more closely with that of the natural images. To this end, we propose to minimize the discrepancy between their respective PF-ODE trajectories from the LR image distribution by our meticulously designed Distribution Trajectory Matching (DTM) loss, resulting in improved realism of our recovered HR images. Comprehensive experimental results demonstrate that the proposed methods can attain comparable or even superior capabilities on both synthetic and real datasets while maintaining minimal inference latency.

Problem

• Eliminates need for multi-step diffusion models in super-resolution
• Directly learns one-step mapping from LR to HR images
• Enhances realism by matching SR and natural image distributions

Innovation

• Distillation-free one-step super-resolution via CTMSR
• PF-ODE trajectory for deterministic LR-to-HR mapping
• DTM loss aligns SR distribution with natural images