Microsoft to release new software for graphic designers: Microsoft has developed a bot that creates images based on text descriptions. For example, he can draw the picture of a bird with a yellow body, black wings and a short beak.The bot can produce images from everyday scenes to the absurd. Interestingly, the images sometimes contain details that are missing in the text descriptions.
“If you’re looking for a bird on Bing, you’ll get a bird photo. But this is where the images are created from scratch, pixel by pixel. These birds may not exist in the real world – they are just the bird’s eye of our computer.”
The network uses two models of machine learning, one that generates images from textual descriptions and one that is known as a discriminator that uses textual descriptions to analyse the authenticity of the images produced.
The generator tries to get fake pictures past the discriminator and on the other hand, does not want to be fooled. This interaction should help to improve the generated images.
The drawing robot was trained on records containing images and image descriptions. In Microsoft’s application, the input text is broken down into individual words and these words are assigned to specific image areas. Because many pictures of birds in the training data show birds sitting on branches, the system usually draws birds that sit on branches, unless the text specifies otherwise.
The team also tried to confuse the drawing robot with nonsensical descriptions. So it was required to draw a red double decker bus floating on a lake. It created a blurry motif that resembles both a boat with two decks and a double-decker bus on a lake surrounded by mountains.
The researchers interpreted the result in such a way that the Bot apparently tried to bring his knowledge that boats float on lakes, in accordance with the text hint on the bus.
Whether there are practical applications for this technique will be revealed. Microsoft believes that text-to-image generation could serve as a sketch wizard for painters and interior designers or as a tool for voice-controlled photo editing. Animated films could be generated with more computational power based on scripts.