New HIT method enables multi-scale image super-resolution

Researchers have developed a new method for multi-scale image super-resolution (ISR) that builds upon Visual Auto-Regressive (VAR) modeling. This approach, called Hierarchical Image Tokenization (HIT), allows for the generation of images at various scales with a single forward pass. It also incorporates Direct Preference Optimization (DPO) regularization to improve performance without requiring extensive external training data or large model backbones. AI

影响 Introduces a more efficient and flexible approach to image super-resolution, potentially improving performance in applications requiring multi-scale outputs.

排序理由 Publication of a new academic paper detailing a novel method for image super-resolution. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CV 阅读 →

• Direct Preference Optimization (DPO)
• Hierarchical Image Tokenization (HIT)
• Visual Auto-Regressive (VAR) modeling

• 论文
• 模型发布

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →