An Intelligent Hybrid Text-To-Image Synthesis Model for Generating Realistic Human Faces
Razan Bayoumi, Marco Alfonse, Abdel-Badeeh Mohamed Salem · 2021
Text-to-image synthesis is referring to converting textual features into pixels, which requires a full understanding of the connection between the natural language text and visual features. What this paper presents is an intelligent hybrid model for converting the textual description into an image; this model depends on the proposed models; AttnGAN and DM-GAN called DMAttn GAN. We focused on applying text-to-image composition or conditional image generation on a new category of the images like human faces. We build a new dataset for text-to-face synthesis which doesn’t get attention because of how challenging this area is as well as the lack of the availability of a suitable dataset. We evaluate the model on faces dataset using the Frechet Inception Distance (FID) with achieving a score of 43.815.