Scholarlink

In order to augment infrared image data and support traffic flow monitoring under low-light conditions, a visible-to-infrared image translation network is proposed in this paper. First, we captured paired visible-infrared traffic image data using dual-modality sensors, helping to address the current lack of publicly available datasets. Second, an enhanced image translation network with Convolutional block attention module (CBAM), Wavelet transform convolution (WTConv) and Atrous spatial pyramid pooling (ASPP) is used for training. Finally, the generated infrared test results are evaluated using Peak Signal-to-Noise Ratio (PSNR), Mean Structural Similarity Index (M-SSIM), and Fréchet Inception Distance (FID), achieving results of 22.1539, 0.7380, and 38.2270. The experimental results show that, compared with the baseline, the proposed network performs well in the task of infrared image generation while maintaining a lower parameter count.

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

SAVI-GAN: An Self-attention Method GAN Using Pix2Pix for Visible to Infrared Image Translation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Xiaoshen Yang,
  • Haoting Liu,
  • Hao Li,
  • Kai Ding,
  • Xiya Chang,
  • Haiguang Li,
  • Xiaoling Ai,
  • Qingwen Hou,
  • Qing Li

摘要

In order to augment infrared image data and support traffic flow monitoring under low-light conditions, a visible-to-infrared image translation network is proposed in this paper. First, we captured paired visible-infrared traffic image data using dual-modality sensors, helping to address the current lack of publicly available datasets. Second, an enhanced image translation network with Convolutional block attention module (CBAM), Wavelet transform convolution (WTConv) and Atrous spatial pyramid pooling (ASPP) is used for training. Finally, the generated infrared test results are evaluated using Peak Signal-to-Noise Ratio (PSNR), Mean Structural Similarity Index (M-SSIM), and Fréchet Inception Distance (FID), achieving results of 22.1539, 0.7380, and 38.2270. The experimental results show that, compared with the baseline, the proposed network performs well in the task of infrared image generation while maintaining a lower parameter count.