Search papers, labs, and topics across Lattice.
This paper introduces TEASR, a novel training-efficient any-step diffusion framework designed for Real-World Image Super-Resolution (Real-ISR), which allows for both one-step and multi-step image restoration without the need for auxiliary teacher models. By implementing self-adversarial distillation and a timestep-aware rectification strategy, TEASR significantly enhances training efficiency, enabling the distillation of large 20B-parameter models on a single GPU. Extensive experiments reveal that TEASR not only improves sampling speed but also consistently outperforms existing state-of-the-art methods across various datasets.
TEASR achieves seamless any-step sampling and outperforms traditional methods, all while eliminating the need for bulky auxiliary models.
Diffusion models excel in Real-World Image Super-Resolution (Real-ISR) due to their powerful generative priors but suffer from slow iterative sampling. Although existing one-step distillation methods accelerate inference, they typically require auxiliary teacher models that inflate training memory and restrict scalability to large-scale architectures. Furthermore, these fixed-step models lack the flexibility to trade off speed for quality. In this paper, we propose TEASR, a training-efficient any-step diffusion framework for Real-ISR that enables both one-step and multi-step restoration within a unified model. Our key idea is to perform self-adversarial distillation within a single diffusion model, eliminating the need for auxiliary teachers or discriminators. Specifically, we propose a timestep-aware rectification strategy that stabilizes one-step generation across noise levels. These two designs further enables the distillation of 20B-parameter diffusion models on a single GPU, significantly improving training efficiency. Moreover, we introduce a dual-branch diffusion transformer with decoupled timestep condition to separate the current noise state and the denoising target to enhance sampling quality. Extensive experiments demonstrate that TEASR supports seamless any-step sampling and consistently outperforms state-of-the-art methods across multiple datasets.
