2.624

2020影响因子

(CJCR)

  • 中文核心
  • EI
  • 中国科技核心
  • Scopus
  • CSCD
  • 英国科学文摘

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

融合自注意力机制和相对鉴别的无监督图像翻译

林泓 任硕 杨益 张杨忆

林泓, 任硕, 杨益, 张杨忆. 融合自注意力机制和相对鉴别的无监督图像翻译. 自动化学报, 2021, 47(9): 2226−2237 doi: 10.16383/j.aas.c190074
引用本文: 林泓, 任硕, 杨益, 张杨忆. 融合自注意力机制和相对鉴别的无监督图像翻译. 自动化学报, 2021, 47(9): 2226−2237 doi: 10.16383/j.aas.c190074
Lin Hong, Ren Shuo, Yang Yi, Zhang Yang-Yi. Unsupervised image-to-image translation with self-attention and relativistic discriminator adversarial networks. Acta Automatica Sinica, 2021, 47(9): 2226−2237 doi: 10.16383/j.aas.c190074
Citation: Lin Hong, Ren Shuo, Yang Yi, Zhang Yang-Yi. Unsupervised image-to-image translation with self-attention and relativistic discriminator adversarial networks. Acta Automatica Sinica, 2021, 47(9): 2226−2237 doi: 10.16383/j.aas.c190074

融合自注意力机制和相对鉴别的无监督图像翻译

doi: 10.16383/j.aas.c190074
详细信息
    作者简介:

    林泓  武汉理工大学计算机科学与技术学院副教授.主要研究方向为图像处理, 数据挖掘, 计算机语言与编译技术.E-mail:linhong@whut.edu.cn

    任硕  武汉理工大学硕士研究生.主要研究方向为计算机视觉和模式识别. E-mail: rensho555@126.com

    张杨忆  武汉理工大学硕士研究生.主要研究方向为计算机视觉和模式识别. E-mail: whutzyy95@163.com

    通讯作者:

    杨益  武汉理工大学计算机学院讲师.主要研究方向为图像处理和模式识别.本文通信作者.E-mail: yang_yi@whut.edu.cn

  • 本文责任编委 白翔

Unsupervised Image-to-Image Translation With Self-Attention and Relativistic Discriminator Adversarial Networks

More Information
    Author Bio:

    LIN Hong   Associate professor at the College of Computer Science and Technology, Wuhan University of Technology. Her research interest covers image processing, data mining, computer language, and compilation technology

    REN Shuo   Master student at Wuhan University of Technology. His research interest covers computer vision and pattern recognition

    ZHANG Yang-Yi  Master student at Wuhan University of Technology. Her research interest covers computer vision and pattern recognition

    Corresponding author: YANG Yi  Lecturer at the College of Computer Science and Technology, Wuhan University of Technology. Her research interest covers image processing and pattern recognition. Corresponding author of this paper
  • Recommended by Associate Editor BAI Xiang
  • 摘要: 无监督图像翻译使用非配对训练数据能够完成图像中对象变换、季节转移、卫星与路网图相互转换等多种图像翻译任务.针对基于生成对抗网络(Generative adversarial network, GAN)的无监督图像翻译中训练过程不稳定、无关域改变较大而导致翻译图像细节模糊、真实性低的问题, 本文基于对偶学习提出一种融合自注意力机制和相对鉴别的无监督图像翻译方法.首先, 生成器引入自注意力机制加强图像生成过程中像素间远近距离的关联关系, 在低、高卷积层间增加跳跃连接, 降低无关图像域特征信息损失.其次, 判别器使用谱规范化防止因鉴别能力突变造成的梯度消失, 增强训练过程中整体模型的稳定性.最后, 在损失函数中基于循环重构增加自我重构一致性约束条件, 专注目标域的转变, 设计相对鉴别对抗损失指导生成器和判别器之间的零和博弈, 完成无监督的图像翻译.在Horse & Zebra、Summer & Winter以及AerialPhoto & Map数据集上的实验结果表明:相较于现有GAN的图像翻译方法, 本文能够建立更真实的图像域映射关系, 提高了生成图像的翻译质量.
    Recommended by Associate Editor BAI Xiang
    1)  本文责任编委 白翔
  • 图  1  模型整体结构

    Fig.  1  The structure of model

    图  2  生成器网络

    Fig.  2  Generator

    图  3  判别器网络

    Fig.  3  Discriminator

    图  4  相对对抗收敛效果

    Fig.  4  Relative discriminator convergence effect

    图  5  本文不同条件实验结果

    Fig.  5  Experimental results under different conditions

    图  6  卫星图与路网图翻译效果对比

    Fig.  6  Comparison of AerialPhoto & Map image translation

    图  7  多种方法翻译效果对比

    Fig.  7  Comparison of multiple image translation methods

    表  1  生成器网络结构参数设置

    Table  1  The parameter setting of generator

    序号 区域划分 层类型 卷积核 步长 深度 归一化 激活函数
    0 下采样 Convolution $ 7 \times 7 $ 1 64 IN ReLU
    1 下采样 Convolution $ 3 \times 3 $ 2 128 IN ReLU
    2 下采样 Convolution $ 3 \times 3 $ 2 256 IN ReLU
    3 中间区 Residual Block $ 3 \times 3 $ 1 256 IN ReLU
    4 中间区 Residual Block $ 3 \times 3 $ 1 256 IN ReLU
    5 中间区 Residual Block $ 3 \times 3 $ 1 256 IN ReLU
    6 中间区 Residual Block $ 3 \times 3 $ 1 256 IN ReLU
    7 中间区 Residual Block $ 3 \times 3 $ 1 256 IN ReLU
    8 中间区 Residual Block $ 3 \times 3 $ 1 256 IN ReLU
    9 上采样 Deconvlution $ 3 \times 3 $ 2 128 IN ReLU
    10 上采样 Self-Attention
    11 上采样 Deconvlution $ 3 \times 3 $ 2 64 IN ReLU
    12 上采样 Convolution $ 7 \times 7 $ 1 3 Tanh
    下载: 导出CSV

    表  2  判别器网络结构参数设置

    Table  2  The parameter setting of discriminator

    序号 层类型 卷积核 步长 深度 归一化 激活函数
    0 Convolution $ 4 \times 4 $ 2 64 LeakyReLU
    1 Convolution $ 4 \times 4 $ 2 128 SN LeakyReLU
    2 Convolution $ 4 \times 4 $ 2 256 SN LeakyReLU
    3 Convolution $ 4 \times 4 $ 2 512 SN LeakyReLU
    4 Convolution $ 4 \times 4 $ 1 1
    下载: 导出CSV

    表  3  本文不同条件分类准确率

    Table  3  CA under different conditions

    数据集 真实图像 相对对抗 自注意力 自注意力+相对对抗
    Horse&Zebra 0.985 0.849 0.862 0.873
    Summer&Winter 0.827 0.665 0.714 0.752
    下载: 导出CSV

    表  4  用户调研评价(%)

    Table  4  User study (%)

    翻译任务 CycleGAN [7] UNIT [8] 本文方法
    Horse $ \rightarrow $ Zebra 10.6 13.3 76.1
    Zebra $ \rightarrow $ Horse 7.6 6.6% 85.8
    Summer $ \rightarrow $ Winter 24.6 19.3% 56.1
    Winter $ \rightarrow $ Summer 15 11.7% 73.3
    下载: 导出CSV

    表  5  分类准确率对比

    Table  5  Classification accuracy comparison

    数据集 真实图像 CycleGAN [7] UNIT [8] 本文方法
    Horse&Zebra 0.985 0.850 0.789 0.873
    Summer&Winter 0.827 0.644 0.591 0.752
    下载: 导出CSV

    表  6  感知距离对比

    Table  6  Perceptual distance comparison

    翻译任务 真实图像 CycleGAN [7] UNIT [8] 本文方法
    Horse $ \rightarrow $ Zebra 1.177 1.133 1.054 1.137
    Zebra $ \rightarrow $ Horse 1.198 1.141 1.056 1.147
    Summer $ \rightarrow $ Winter 1.824 1.189 1.153 1.211
    Winter $ \rightarrow $ Summer 1.272 1.223 1.209 1.259
    下载: 导出CSV
  • [1] Isola P, Zhu J Y, Zhou T H, Efros A A. Image-to-image translation with conditional adversarial networks. In: Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, Hawaii, USA: IEEE, 2017. 1125-1134
    [2] Xu F R, Ma B P, Chang H, Shan S G, Chen X L. Style transfer with adversarial learning for cross-dataset person re-identification. In: Proceedings of the 2018 Asian Conference on Computer Vision. Perth, Australia: Springer, 2018.
    [3] Goodfellow I J, Pouget-Abadie J, Mirza M, Xu B, WardeFarley D, Ozair S, et al. Generative adversarial nets. In: Proceedings of the 27th International Conference on Neural Information Processing Systems. Montreal, Canada: MIT Press, 2014. 2672-2680
    [4] Huang H, Yu P S, Wang C. An introduction to image synthesis with generative adversarial nets. arXiv Preprint arXiv: 1803.04469, 2018.
    [5] Mirza M, Osindero S. Conditional generative adversarial nets. arXiv preprint arXiv: 1411.1784, 2014.
    [6] Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation. In: Proceedings of the 18th International Conference on Medical Image Computing and Computer-assisted Intervention. Munich, Germany: Springer, 2015. 234-241
    [7] Zhu J Y, Park T, Isola P, Efros A A. Unpaired image-toimage translation using cycle-consistent adversarial networks. In: Proceedings of the 2017 IEEE International Conference on Computer Vision. Venice, Italy: IEEE, 2017. 2223 -2232
    [8] Liu M Y, Breuel T, Kautz J. Unsupervised image-to-image translation networks. In: Proceedings of the 31st Conference on Neural Information Processing Systems. Long Beach, California, USA: MIT Press, 2017. 700-708
    [9] Miyato T, Kataoka T, Koyama M, Yoshida Y. Spectral normalization for generative adversarial networks. arXiv Preprint arXiv: 1802.05957, 2018.
    [10] He D, Xia Y C, Qin T, Wang L W, Yu N H, Liu T Y, et al. Dual learning for machine translation. In: Proceedings of the 30th Conference on Neural Information Processing Systems. Barcelona, Spain: NIPS, 2016. 820-828
    [11] Johnson J, Alahi A, Li F F. Perceptual losses for real-time style transfer and super-resolution. In: Proceedings of the 2016 European Conference on Computer Vision. Amsterdam, The Netherlands: Springer, 2016. 694-711
    [12] He K M, Zhang X Y, Ren S Q, Sun J. Deep residual learning for image recognition. In: Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, NV, USA: IEEE, 2016. 770-778
    [13] Ulyanov D, Vedaldi A, Lempitsky V. Instance normalization: The missing ingredient for fast stylization. arXiv Preprint arXiv: 1607.08022, 2016.
    [14] Jolicoeur-Martineau A. The relativistic discriminator: A key element missing from standard GAN. arXiv Preprint arXiv: 1807.00734, 2018.
    [15] Mao X D, Li Q, Xie H R, Lau R Y, Wang Z, Smolley S P. Least squares generative adversarial networks. In: Proceedings of the 2017 IEEE International Conference on Computer Vision. Venice, Italy: IEEE, 2017. 2794-2802
    [16] Russakovsky O, Deng J, Su H, Krause J, Satheesh S, Ma S, et al. Imagenet large scale visual recognition challenge. International Journal of Computer Vision, 2015, 115(3): 211-252
    [17] Heusel M, Ramsauer H, Unterthiner T, Nessler B, Hochreiter S. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In: Proceedings of the 31st Conference on Neural Information Processing Systems. Long Beach, California, USA: MIT Press, 2017. 6626-6637
    [18] Kingma D P, Ba J. Adam: A method for stochastic optimization. arXiv Preprint arXiv: 1412.6980, 2014.
    [19] Chollet F. Xception: Deep learning with depthwise separable convolutions. In: Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, Hawaii, USA: IEEE, 2017. 1251-1258
    [20] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition. arXiv Preprint arXiv: 1409.1556, 2014.
  • 加载中
图(7) / 表(6)
计量
  • 文章访问数:  188
  • HTML全文浏览量:  68
  • PDF下载量:  87
  • 被引次数: 0
出版历程
  • 收稿日期:  2019-01-29
  • 录用日期:  2019-05-19
  • 刊出日期:  2021-10-13

目录

    /

    返回文章
    返回