Text-to-Image Synthesis using stackGAN
Y. Lakshmi Sahithi, N. Sunny, Marballi Deepak, Amrutha S. V. · 2023
In text-to-image conversion, it can be challenging to produce visuals of a high caliber from written descriptions. Examples created using traditional text to image approaches can usually convey the meaning of the descriptions given, but they are deficient in important details and vibrant object aspects. The creation of photo-realistic graphics from text has several applications, including photo editing, computer-aided design, gaming, virtual reality, and medical imaging. This project suggests using StackGANs, or stacked generative adversarial networks, to create accurate images depending on the text given. There are two phases to StackGAN. Stage-I GAN generates low-resolution representations of the object by sketching its basic form and colors from the text provided. The Stage-II GAN uses text descriptions and Stage-I discoveries as inputs to create high-resolution images with lifelike qualities.