Researchers have developed CoScan, a novel content-adaptive scanning method for causal vision State Space Models (SSMs) designed to improve image restoration tasks. Unlike existing fixed scan orders, CoScan learns image-dependent traversals by encouraging short-range coherence and causal predictability. The method incorporates an efficient multi-scale scan construction and a frequency-guided start selection strategy to enhance causal context formation. When integrated with Mamba-based backbones, CoScan demonstrates consistent improvements across various restoration benchmarks with minimal computational overhead. AI
IMPACT Enhances image restoration capabilities by improving the efficiency and effectiveness of vision State Space Models.
RANK_REASON The cluster contains a research paper detailing a new method for image restoration using vision SSMs. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →