What Inspired Me:

I was really impressed by the idea that machines could create new and original works of art when I first saw Google's Deep Dream back in 2015. It was so novel at the time, and it's what got me into Machine Learning. I then later on learnt about GANs, but the task looked daunting didn't know where to start, as a highschooler. Being part of this boot-camp gave me the courage and mentorship to start.

What I Learned:

First of all, I learnt a lot about Python, this is the first large scale project I have done in Python, and I think it increased my proficiency by a lot. I then learned a lot about reading data, and data pre-processing, knowing how to choose and work with a dataset was a much more important step than I thought. Finally, I learnt a lot about Machine Learning, specifically the tensorflow platform, and Generative Adversarial Networks in general, and it could include Classifiers and Convolutional Neural Networks too, since the GAN's Generator and Discriminator are basically those things.

How I Built My Project:

I used the Stanford Dog Dataset, and used a DCGAN, along with Spectral Normalization. The webapp is made using Flask, and React

Challenges Faced:

The hardest challenge was building the GAN, period. It was so hard to tune the parameters, and it took a lot of time to train, thus giving me less time to try more settings with more iterations. I had to do a lot of reading to get a good understanding of what's going on, and even then, the results are not that amazing.

Built With

  • jupyter-notebook
Share this project:

Updates