Vastr-GAN: Versatile Apparel Synthesised from Text using a Robust Generative Adversarial Network

The development of the fashion industry has increased the demand for customised and meticulously designed clothes. This poses a challenge to fashion designers who need to create novel clothing designs based on the requirements specified by the customers. This work presents a generative adversarial network (GAN) based text-to-image synthesis model for fabricating intricate Indian apparel designs. We introduce an architecture that strategically combines multiple trained GAN models for a streamlined text-to-image generation. Existing fashion datasets with elaborate image descriptions cater to western fashion only. We have extracted traditional Indian images like kurtis, kurtas, etc., and then combined with an existing dataset to create an Indian Fashion dataset of around 16000 images with their corresponding text descriptions. On carrying out elaborate testing on our dataset we have achieved good visual results that can capture the details given in the text descriptions.