Inspiration
Saudi Arabia has different regions. Each region has its own unique heritage. Our diverse heritage can be seen in the food, costumes, dialects and so on. But visually & aesthetically, architecture is the most interesting aspect of any heritage. Unfortunately, the media tends to focus mainly on the Saudi natural heritage, neglecting other important aspects of our culture like the architectural heritage. And there is no better way to describe it than using colors and movement.
What it does
-Tells the story of our rich heritage and diverse society through abstract colors & movement, using digital illustrations inspired from the Saudi architectural heritage as dataset.
How we built it
- Our own advanced GAN model
- Latent space
- A dataset of original illustrations made from scratch -Using color swap method to increase dataset
- Train model to increase resolution of images
Challenges we ran into
-Create GAN model that generate rectangle image -Creating a large number of original art (illustrations) as dataset -Exploring latent space interpolation -Experimenting on a different kind of GAN model
Accomplishments that we're proud of
-Creating a large number of original art (illustrations) as dataset in a limited period of time Modified PyTorch implementation of WGAN so we can artwork that will fill correctly into screen size -Before getting the full number of dataset, training GAN model with a small number of dataset and getting acceptable results was an accomplishment -Train algorithms to avoid mode of collapse
- Made alot of experiments on many type of GAN models
What we learned
-We learned a lot about the history of Saudi architectural heritage -How to choose the best GAN for our dataset -Making correct data augmentation to get the best results -We learned about latent space
What's next for 43_The Beauty of Belonging - Initial Project Submission
Research how to transform the idea into an “experience”.
Log in or sign up for Devpost to join the conversation.