Manifold Matching Generative Adversarial Network (MM-GAN)

Introduction:

This was a special project as I got to work related to Generative Adversarial Networks(GAN). I was responsible for building a web application which could help admin upload sets of real and fake images from various algorithms and the user could perform the test to identify fake images. The test includes endless steps and each step had a grid of some random fake and real images which are displayed to the user and he had to identify the correct fake image and mark down that image. If the user ends the test, the final result is displayed and same are sent to admin. The development was done in Python using Django as a framework.

Major Functionalities:

• Unique passkey generation for each test taker.
• Bulk image upload upto 100k in a batch of 5000.
• Endless test steps.
• Result calculation.

Technical Sheet:

• Python & Django
• Task queueing using Celery and Redis.
• Form submission using Ajax.
• HTML, CSS and JavaScript.