Text Description to Facial Sketch Generation using GANs

Text-to-face generation is an exciting area of research focused on generating human faces based on textual descriptions, presenting unique challenges that have primarily been explored in academia. The advancement of generative adversarial networks (GANs) has played a crucial role in implementing text-to-face generation. This study aims to develop a framework that allows users to express their desired human face characteristics through textual descriptions, with the tool generating lifelike human face images in response. The approach utilizes a Deep Fusion generative adversarial network (DFGAN) architecture, incorporating deep convolutional neural networks (CNNs), trained on the CelebA dataset, which provides a diverse collection of labeled face images for training. The two model's performance are evaluated using the Fréchet Inception Distance (FID) score. DFGAN has achieved a FID score of 100 after evaluating 15k images, surpassing the results obtained by the DCGAN model.