Generative adversarial network for image generation

We developed a generative model (GAN) capable of creating high-quality Street View House Numbers (SVHN) images along with corresponding accurate labels. The generative model output will be used as the input data for a fixed classifier training.

To be specific, we implemented a Conditional DCGAN. However, due to time constraints I would still like to experiment using Attention layers instead of convolutional layers following one of the papers I have read. (I will try this out soon)

We ended up being first place and qualified for the European round.