Text-to-Face Generation Using DCGAN with Bert-Embedding Vectors

The aim of text-to-face generation is to create face images from textual facial descriptions. It has a big impact on a lot of different public safety applications as well as new research sectors. Because there aren’t enough datasets, there isn’t much research done on text-to-face generation. This research uses deep convolutional generative adversarial networks with Bert embedding models to improve text-to-face image synthesis. Methodology incorporates domain-specific BERT embedding models for better understanding of face text descriptions. To enhance its knowledge understanding for the face domain, the BERT model is fine-tuned over large text face description corpus data as a transfer learning method, subsequently enhancing word embedding for substantial textual data. This paper presents an improved Bert model training approach with DCGAN for facial feature understanding and face image generation. The evaluation of the results is in terms of binary cross entropy loss over a period of epochs and the realistic image outputs generated by the approach model.