Manav Doshi & Kaivaly Daga
8-10
Have you ever wondered what you and your (celebrity) crush's kid will look like? Find out now by completing your SoC project.
Explore different Generative Adversarial Networks (GANs), which have been coined as the most exciting idea in machine learning in the past decade. These networks have limitless applications, from generating new celebrities' images to automated 3D object generation. We will convert parents' images to predict their child's image using FamilyGAN.
In this project, we will implement various types of Convolutional Neural Networks(CNNs) for image processing and recognition and finally implement a Generative Adversarial Network(GAN) to predict childrens' faces from their parents.
This is similar how "FaceApp" works. Our training dataset will consist of images of parents and their offspring. GAN is a kind of neural network that will be perfect for such tasks.
I can surely think of a few applications for this :p. Deploying this on a web app would be a plus point, after all who knows if this website can be a hit on campus?
References:
https://medium.com/swlh/familygan-generating-a-childs-face-using-his-parents-394d8face6a4
http://cs230.stanford.edu/projects_fall_2019/reports/26256603.pdf
https://github.com/munozalexander/Child-Face-Generation
Prerequisites:
Enthusiasm is required. Basics of Python are a bonus.
The project could be a little heavy if you were to explore multiple methods.
Week | Work |
---|---|
Week 1 | Learning basics of GitHub and python. |
Week 2 | Deep learning course of Coursera |
Week 3 | Building a basic CNN |
Week 4 | Literature review and setting up the problem |
Week 5-6 | Implementation of the model and testing it using the data |
Week 7 | Fix all errors and complete the documentation |