使用DCGAN结构结合WGAN损失函数在CelebA数据集上训练生成人脸

• 训练数据 CelebA数据集 http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html

  • 解压到./img_align_celeba文件夹下
  • ./m_img: 存放中间图片
  • ./ckpt: 存放中间权重

• 原始GAN训练

  

• WGAN训练

  

• 原始GAN训练100个epoch

  

​

• WGAN训练300个epoch

  

分别是第0，100，200，299个epoch时生成器G输出的结果

参考资料

[1] https://github.com/pytorch/examples/blob/main/dcgan/main.py

[2] https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html

[3] https://blog.csdn.net/qq_37395293/article/details/103112721

[4] Goodfellow I, Pouget-Abadie J, Mirza M, et al. Generative adversarial nets[J]. Advances in neural information processing systems, 2014, 27.

[5] Radford A, Metz L, Chintala S. Unsupervised representation learning with deep convolutional generative adversarial networks[J]. arXiv preprint arXiv:1511.06434, 2015.

[6] Arjovsky M, Chintala S, Bottou L. Wasserstein generative adversarial networks[C]//International conference on machine learning. PMLR, 2017: 214-223.