Conference Proceedings
The International Conference on Emerging Technologies in Electronics, Computing and Communication 2022
(ICETECC`22)
Text-to-Image Generation Using Recurrent Convolutional GANs
Hafiz Arslan Ramzan1*; Sadia Ramzan2; Tehmina Kalsum3; Minahil Rauf1; Maria Sher4;| 1NUST 2NUCES,Faislabad 3UET Taxila 4KFUEIT |
ABSTRACT
Text-to-image creation has become a major area of interest for Generative Adversarial Networks (GANs) due to their recent advancements. The main objective of this study is to generate realistic images from textual descriptions using a novel method that combines convolutional GANs with recurrent neural networks (RNNs). The suggested model, known as Recurrent Convolutional GAN (RCGAN), makes use of both the powerful image generation abilities of convolutional GANs and the sequential character of RNNs to capture complicated textual semantics. By obtaining an Inception Score of 4.25 and a Peak Signal-to-Noise Ratio (PSNR) of 31.07 dB on the Oxford-102 Flowers dataset, we demonstrate the efficiency of RCGAN. The results suggest that RCGANs offer an exciting potential for further generative model research because they perform more effectively than current techniques in terms of textual relevance and visual fidelity.