Scholarlink

<p>The paper introduces a new method for image super-resolution (SR) using a prompt-guided, zero-shot framework that leverages the capabilities of vision-language foundation models (VLFMs) and generative diffusion models. Unlike traditional SR models requiring paired low and high-resolution images, this approach uses natural language prompts to enhance images without needing paired data. VLFMs like BLIP help create strong representations by connecting low-resolution images with various texts. The framework employs static and dynamic prompt techniques to adapt to different image contexts and user needs. Tests on datasets such as DIV2K confirm its effectiveness, showing it achieves performance on par with or better than conventional methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Zero-shot image super-resolution using prompt-driven vision-language foundation models without task-specific fine-tuning

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • K. Akila

摘要

The paper introduces a new method for image super-resolution (SR) using a prompt-guided, zero-shot framework that leverages the capabilities of vision-language foundation models (VLFMs) and generative diffusion models. Unlike traditional SR models requiring paired low and high-resolution images, this approach uses natural language prompts to enhance images without needing paired data. VLFMs like BLIP help create strong representations by connecting low-resolution images with various texts. The framework employs static and dynamic prompt techniques to adapt to different image contexts and user needs. Tests on datasets such as DIV2K confirm its effectiveness, showing it achieves performance on par with or better than conventional methods.