pytorch-MNIST-CelebA-GAN-DCGAN

Pytorch implementation of Iterative Knowledge Distillation for Improved Generative Model Compression based on Deep Convolutional Generative Adversarial Networks (DCGAN) [1] for MNIST [2] datasets.

• you can download
  • MNIST dataset: http://yann.lecun.com/exdb/mnist/

Implementation details

• DCGAN

Resutls

• Original model vs Compressed model

• FID

  • 128-64
  • 128-95
  • 95-64

• FLOP

  • 128-64 - 1.646G
  • 128-95 - 913.792M
  • 95-64 - 420.573M

Development Environment

• Ubuntu 16.04 LTS
• NVIDIA GTX 1080 ti
• cuda 10.1
• Python 3.7.3
• pytorch 1.2.0
• torchvision 0.1.8
• matplotlib 3.1.0
• imageio 2.5.0
• scipy 1.3.0
• pillow 6.1.0
• scikit-image 0.15.0

Reference

[1] Radford, Alec, Luke Metz, and Soumith Chintala. "Unsupervised representation learning with deep convolutional generative adversarial networks." arXiv preprint arXiv:1511.06434 (2015).

(Full paper: https://arxiv.org/pdf/1511.06434.pdf)

[2] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. "Gradient-based learning applied to document recognition." Proceedings of the IEEE, 86(11):2278-2324, November 1998.