Scholarlink

<p>In recent years, generative adversarial networks (GANs) have demonstrated enormous promise in areas connected to image generation. As the model generation performance continues to improve and the generated images become more realistic, it is difficult to effectively distinguish between the real image and the generated image. Therefore, the problem of discriminating and optimizing the generated images (adversarial discrimination) has become necessary, and subsequent optimization plans are proposed based on the discrimination strategy. However, due to the nature of convolution, the two-dimensional power spectrum curve of the generated image is low overall; that is, compared with the real image, there is energy loss at each frequency (without other processing), and the curve drops rapidly and approaches zero, which is obviously different from the real image. In particular, the curve of the image generated by transposed convolution has a clear upward trend at the very high-frequency part, which is contrary to the characteristic of the real image, which is that the energy decreases with increasing frequency. By analyzing the characteristics and causes of the two-dimensional power spectrum curve of generated images, we propose a discrimination approach that leverages high-frequency curve warping and energy loss. This approach enhances the ability to discriminate generated images and enables effective distinction between real and generated images. Based on this, we present the power spectrum loss function to improve the upward warping characteristics of the very high-frequency part of the two-dimensional power spectrum curve without degrading the quality of the generated image and the high-frequency feature loss function to improve the quality of the generated image. The value and efficiency of the proposed discrimination approach in this study are demonstrated on multiple GANs models, including WGAN, WGAN-GP, and SAGAN, with the dataset CelebA, and the GANs model with encoder-decoder as the generator with the dataset CelebA-HQ. The two loss functions proposed are also demonstrated on multiple GANs models, including WGAN, WGAN-GP, and SAGAN with the dataset FFHQ. After adding the high-frequency feature loss, the Fréchet inception distance (FID) decreases by 6.97, 6.15, and 5.56, respectively. After adding the power spectrum loss, the above models can improve the upward warping characteristics of the two-dimensional power spectrum curve in the very high-frequency part of the generated image to a certain extent. The FID decreases by 17.4, 11.55, and 12.27 when the weight is fixed, and 12.66, 8.15, and 4.46 when the weight is variable, respectively.</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

Frequency domain discrimination and double loss function optimization of image generation for GANs

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月16日

查看全文
  • Xueyi Ye,
  • Mingcong Sui,
  • Maosheng Zeng,
  • Zhuo Han,
  • Hao Wang

摘要

In recent years, generative adversarial networks (GANs) have demonstrated enormous promise in areas connected to image generation. As the model generation performance continues to improve and the generated images become more realistic, it is difficult to effectively distinguish between the real image and the generated image. Therefore, the problem of discriminating and optimizing the generated images (adversarial discrimination) has become necessary, and subsequent optimization plans are proposed based on the discrimination strategy. However, due to the nature of convolution, the two-dimensional power spectrum curve of the generated image is low overall; that is, compared with the real image, there is energy loss at each frequency (without other processing), and the curve drops rapidly and approaches zero, which is obviously different from the real image. In particular, the curve of the image generated by transposed convolution has a clear upward trend at the very high-frequency part, which is contrary to the characteristic of the real image, which is that the energy decreases with increasing frequency. By analyzing the characteristics and causes of the two-dimensional power spectrum curve of generated images, we propose a discrimination approach that leverages high-frequency curve warping and energy loss. This approach enhances the ability to discriminate generated images and enables effective distinction between real and generated images. Based on this, we present the power spectrum loss function to improve the upward warping characteristics of the very high-frequency part of the two-dimensional power spectrum curve without degrading the quality of the generated image and the high-frequency feature loss function to improve the quality of the generated image. The value and efficiency of the proposed discrimination approach in this study are demonstrated on multiple GANs models, including WGAN, WGAN-GP, and SAGAN, with the dataset CelebA, and the GANs model with encoder-decoder as the generator with the dataset CelebA-HQ. The two loss functions proposed are also demonstrated on multiple GANs models, including WGAN, WGAN-GP, and SAGAN with the dataset FFHQ. After adding the high-frequency feature loss, the Fréchet inception distance (FID) decreases by 6.97, 6.15, and 5.56, respectively. After adding the power spectrum loss, the above models can improve the upward warping characteristics of the two-dimensional power spectrum curve in the very high-frequency part of the generated image to a certain extent. The FID decreases by 17.4, 11.55, and 12.27 when the weight is fixed, and 12.66, 8.15, and 4.46 when the weight is variable, respectively.