Short Bytes: StacKGAN is a new AI in the house which is capable of creating photo-realistic images out of plain text. Developed at the Rutgers University, the system generates high-quality 256×256 pictures following a two-stage process after analyzing the given text description.The researchers at the Rutgers University’s Computer Science department have created an AI system which does the opposite of the task performed by Google’s Show and Tell. Known as StackGAN, the neural network is capable of generating a high-quality image after analyzing the text description of the same. Researchers from the Lehigh University, The Chinese University of Hong Kong, and the University of North Carolina have also contributed to the development of this AI.
Stack General Adversarial Networks can process description and create an output image after a two-stage procedure. The first stage involves the creation of the basic image along with filling appropriate colors. The second step adds the necessary refinements to image generated in the first stage.
The team notes the fact that other text-to-image methods exist. But, StackGAN supersedes others in terms of picture quality and creates photo-realistic images with 256 x 256 pixels in comparison to 128×128 for others.
Such AI system, if improved further, can have many applications. For instance, it can help in police investigations. Photo illustrations of people involved in the case are created from their description.
The new artificial intelligence system adds to the list of impressive AI’s we have seen in 2016. That includes the tv watching AI created by Deepmind, Google Brain’s AI which uses its own language to translate unknown language pairs, and IBM Watson which created the trailer of the movie Morgan.
For detailed information, you can read the research paper StackGAN: Text to Photo-Realistic Image Synthesis with Stacked Generative Adversarial Networks
If you have something to add, tell us in the comments below.