swapnilbembde / alpha-numeric_data_generation_using_GAN

Pytorch implementation of Generative Adversarial Networks (GAN) for MNIST and EMNIST datasets

Data generation using random noise

Introduction

This project is a part of our course work (EE769). This project aims at expanding any current dataset by generating similar images using GANs concept. This expanded dataset will contribute to other machine learning problems. We take random noise as an input and transform into usable handwriting texts.

Results

MNIST	EMNIST

Development Environment

• Ubuntu 16.04 LTS
• NVIDIA 840M
• cuda 9.0
• Python 2.7.6
• pytorch 0.4.0
• torchvision 0.1.4
• matplotlib 1.3.1
• imageio 2.2.0
• scipy 0.19.1

References

Goodfellow, Ian, et al. "Generative adversarial nets." Advances in neural information processing systems. 2014. (http://papers.nips.cc/paper/5423-generative-adversarial-nets.pdf)

Our code is inspired by: https://github.com/znxlwm/pytorch-generative-model-collections