Multi Agent Diverse Generative Adversarial Network (MAD-GAN)

Pytorch Code for MAD-GAN.

Requirements

• pytorch >=0.4.0
• torchvision ==0.2.0
• Jupyter Notebook

Datset

• A distribution of 1D GMM having five mixture components with modes at 10, 20, 60, 80 and 110, and standard deviations of 3, 3, 2, 2 and 1, respectively.

Usage

• Run the Simple_GANs.ipynb to generate the results for vanilla GAN.
• For MAD-GAN run the Mad_GANs.ipynb to compare the results of vanillaGAN over MAD-GAN.
• Number of generater of MAD-GAN can by changing 'num_gen' and G.params().

Results

             [Left]: Result of VanillaGAN                [Right]: Result of MAD-GAN (number of generator = 4)

Resources

• Multi-Agent Diverse Generative Adversarial Networks

• Ghosh, Arnab, et al. "Multi-agent diverse generative adversarial networks." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018.

Acknowledgements

• https://github.com/wiseodd
• Ghosh, Arnab, et al. "Multi-agent diverse generative adversarial networks." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018