Scholarlink

This paper explores the use of Generative Adversarial Networks (GANs) for realistic texture synthesis in style transfer tasks. The study utilizes StyleGAN for high-resolution texture generation and CycleGAN for bidirectional style translation. To optimize adversarial losses, content, and style, the models undergo extensive training for 300 and 2200 epochs, respectively. The results are evaluated using the Structural Similarity Index (SSIM) and Fréchet Inception Distance (FID), along with visual inspection. Using the proposed approach, one can easily realize that the architectures of these GANs can provide efficient solutions for texture synthesis and style transfer.

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

Generative Adversarial Networks for Realistic Texture Synthesis in Style Transfer

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Shivam Kumar,
  • Kulvinder Singh,
  • Tanu Dahiya,
  • Mukul Shukla,
  • Subham Kumar Mishra,
  • Krish Sinha

摘要

This paper explores the use of Generative Adversarial Networks (GANs) for realistic texture synthesis in style transfer tasks. The study utilizes StyleGAN for high-resolution texture generation and CycleGAN for bidirectional style translation. To optimize adversarial losses, content, and style, the models undergo extensive training for 300 and 2200 epochs, respectively. The results are evaluated using the Structural Similarity Index (SSIM) and Fréchet Inception Distance (FID), along with visual inspection. Using the proposed approach, one can easily realize that the architectures of these GANs can provide efficient solutions for texture synthesis and style transfer.