Image Generation Using VQ-GAN, CLIP, and Custom Deep Face GAN
Team: Beta Learners
Developed as part of Hack-A-Roo and CSEE-5590-0001-13590-2022SP-Special Topics - Python/Deep Learning Class University of Missouri - Kansas City
Problem Statement: Image generation is a critical step for Artificial Intelligence technologies in understanding image properties at the pixel level concerning context and environment. Generative Adversarial networks along with transformers played a crucial role in generating the image content with specified parameters. The main idea presented in this paper is to develop deep learning models to take the celebrity dataset related to face images and generate similar images related to technically fake faces. Two models are designed to handle the solution for deep face image generation and text to image using VQ-GAN and CLIP. Realistic than the original images with and without context. Taking input from the user like image type and context related to the image and generating the images at a pixel level. Both are different models as they need separate training and development. These generated images are used in a wide variety of multi-media, Non-Fungible Token (NFT) developers, UI/UX designers, developers across the industries of animation and internet companies, and research groups for generating data.
Youtube Link: https://youtu.be/rRnQODwcZN4