Text to Photo-Realistic Image Synthesis using Generative Adversarial Networks
Abhishek S. Rao, P Adokshaj Bhandarkar, Padmashali Akshay Devanand, Pratheek Shankar, Srinivas Shanti, Karthik Pai B H · 2023
Innovative design patterns have traditionally been constrained by time-consuming and expertise-intensive methods, limiting diversity and creativity in design. To overcome these challenges, our research focuses on revolutionizing the design process by leveraging Generative Adversarial Networks (GANs). Our goal is to offer an affordable and cutting-edge solution that enables the creation of unique design patterns based on user input text. By harnessing the power of GANs, we can generate high-resolution images, reducing the reliance on expensive designers and fostering creativity and idea generation in the design sector. Our methodology involves building upon a diverse dataset of sunflower images and associated text data borrowed from the Oxford-102 flowers dataset. We meticulously collect, clean, and normalize this dataset to ensure optimal training. U sing an encoder-decoder architecture, we train a GAN -based generative model to produce remarkable sunflower patterns. To enhance usability, we have developed a user-friendly graphical user interface (GUI) that allows users to input text and receive corresponding sunflower images seamlessly. Our current model has exhibited promising results, achieving an inception score of 3.45 ± 0.05 in generating sunflower patterns. These outcomes demonstrate the feasibility of our approach, which not only reduces design costs but also fosters innovation and ingenuity within the industry. By making this solution publicly accessible, we aim to support local businesses and individuals with limited finances in creating distinctive design patterns.