Multi-Sentence Hierarchical Generative Adversarial Network GAN (MSH-GAN) for Automatic Text-to-Image Generation

This research work is related to the development of technology in the field of automatic-text-to-image generation. In this regard, two main goals are pursued. First, the generated image should look as real as possible, and secondly, the generated image should be a meaningful description of the input text. Our proposed method is a multi-sentence hierarchical generative adversarial network (MSH-GAN) for the text-to-image generation. In this research project, we consider two main strategies: 1) produce a higher quality image in the first step, and 2) use two additional descriptions in order to improve the original image in the next steps. Our goal is to focus on using more information to generate images with a higher resolution using more than one sentence input text. We propose different models based on GANs and memory networks. We also use a more challenging dataset called ids-ade. This is the first time; this dataset has been used in this area. We evaluate our models based on the IS, FID, and R-precision evaluation metrics. The experimental results obtained demonstrate that our best model performs favorably against the basic state-of-the-art approaches like StackGAN and AttGAN.