Scholarlink

Recent researches in deep neural networks have the ability to create different styles of high quality music, which is comparable to music composed by humans. However, few work enables music generation from images. In this paper, we present ImageDJ, a generative model designed to compose music based on specific images. We use an image classification model to categorize images according to various styles and create a style embedding that guides the process of music generation. Our evaluation result shows that ImageDJ can generate high quality musical compositions inspired by various styles captured in images.

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

Style-Specific Music Generation from Image

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Chang Xu,
  • Xuan Liu,
  • Yu Weng,
  • Shan Jiang,
  • Xiangyun Tang,
  • Minfeng Qi

摘要

Recent researches in deep neural networks have the ability to create different styles of high quality music, which is comparable to music composed by humans. However, few work enables music generation from images. In this paper, we present ImageDJ, a generative model designed to compose music based on specific images. We use an image classification model to categorize images according to various styles and create a style embedding that guides the process of music generation. Our evaluation result shows that ImageDJ can generate high quality musical compositions inspired by various styles captured in images.