New framework Allo{SR}$^2$ enhances one-step image super-resolution

Researchers have introduced Allo{SR}$^2$, a new framework designed to improve one-step image super-resolution (Real-SR) by addressing distribution shifts and trajectory deviations common in existing methods. The framework utilizes SNR-Guided Trajectory Initialization to align low-resolution representations with generative flows and employs Flow-Anchored Trajectory Consistency to stabilize inference paths. Additionally, Allomorphic Trajectory Matching is used to preserve generative realism, leading to state-of-the-art performance in one-step Real-SR with enhanced efficiency. AI

IMPACT This research could lead to more efficient and realistic image upscaling in various applications.

RANK_REASON This is a research paper detailing a new technical framework for image super-resolution. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Allo{SR}$^2$
• arXiv
• Diffusion Models
• Flow Matching
• Hugging Face
• Zihan Wang

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →