Research Article
Trends in Text to Image Synthesis (T2I) using Generative Adversarial Networks
@INPROCEEDINGS{10.4108/eai.7-12-2021.2315122, author={Venkatesan R and Priyanka S}, title={Trends in Text to Image Synthesis (T2I) using Generative Adversarial Networks}, proceedings={Proceedings of the First International Conference on Combinatorial and Optimization, ICCAP 2021, December 7-8 2021, Chennai, India}, publisher={EAI}, proceedings_a={ICCAP}, year={2021}, month={12}, keywords={generative adversarial networks deep attentional multi-layered natural language textual descriptions photo-realistic semantic consistency text-to-image synthesis}, doi={10.4108/eai.7-12-2021.2315122} }- Venkatesan R
Priyanka S
Year: 2021
Trends in Text to Image Synthesis (T2I) using Generative Adversarial Networks
ICCAP
EAI
DOI: 10.4108/eai.7-12-2021.2315122
Abstract
The two most evident modalities of humans are language and vision. Any system that aids interaction between human beings and Artificial Intelligence (AI) is rooted upon these two. Text-to-Image synthesis (T2I) powered by Natural Language Processing (NLP) and deep Generative Adversarial Networks (GANs) replicates this phenomenon. The logical relationship between semantics and vision guides T2I, that attempts to translate highly detailed natural language textual descriptions to pixel-level details. The human concept of attention is leveraged and conceptualized by deep attentional multi-layered GANs. Mimicking the human thinking processes of visualizing the scenes in mind while speaking and listening can be extensively used in various AI applications that craves brain-like comprehending potency. The advancement of a multitude of GANs that focused on semantic consistency, high-resolution photo-realistic images and diversity in synthesis has been investigated in this article.