Text to Image Synthesis using Residual GAN
Priyanka P. Mishra, Tribhuvan Singh Rathore, Shivani Shivani, Sachin Tendulkar · 2020
In the world of computer vision, a very intriguing problem is synthesizing or generating images (from the noise) of the reasonable quality from text descriptions. The applications of this problem are immense such as photo-editing, computer- aided design, etc. But the current AI systems are not up to the mark to reach the desired outcome. However, in recent years the progress in the field of text classification and image classification fields have paved the way for more advanced AI systems that can be used to achieve the desired goal by utilizing the discriminative power and strong generalization properties of attribute representations of recurrent neural networks and convolutional neural networks. Meanwhile, GANs have proved to produce reasonable images of birds, flowers, etc. In this work, we present GAN architecture to effectively aid the translation visual concepts from the text to image.