Automation of Forensic Artist in Criminal Investigation Using Generalized Adversarial Networks

EasyChair Preprint 10641

Abstract

In recent years, AI-driven picture production has greatly advanced. Generative Adversarial Networks (GANs), such as the Style GAN, can provide realistic data of the highest calibre while also allowing for creative input. In order to create a detailed human face from textual description, we describe a way of managing text output in this study. We modify various face aspects using Style GAN's latent space and conditionally sample the necessary latent code, which embeds the facial features specified in the input text. Our approach demonstrates accurate feature capture and demonstrates consistency between the input text and the output photos. Additionally, our approach ensures disentanglement while changing a variety of facial traits that adequately represent a human face.

Keyphrases: CLIP, Discriminator, Generative Adversarial Networks (GAN), generator

@booklet{EasyChair:10641,
  author    = {S.K Prashanth and Roopa Sri Gaddam and Pulluri Chandana},
  title     = {Automation of Forensic Artist in Criminal Investigation Using Generalized Adversarial Networks},
  howpublished = {EasyChair Preprint 10641},
  year      = {EasyChair, 2023}}