Microsoft gave a textual description, and the AI drew a bird
Already aware of the algorithms that can determine what is depicted in the photo. Usually AI will recognize such photos with high accuracy (up to 80%). Now, Microsoft has put a new bar before artificial intelligence. Researchers described in words what they want to see in the image. Namely – a yellow bird with black wings and a short beak. AI painted a bird, and very accurate!
The picture looks like a real photo. And this is all the more impressive, because the image was not created on the basis of others. The finished database was not used. The algorithm created the drawing itself.
Xiaodong He, who led the research, says that this is not like when you enter a word in the search string, and get the appropriate images. Here, the photo is created by the AI on its own. The image is collected by pixels. That is, this is not yet. Perhaps, the depicted in general does not exist in nature. It’s like the imagination of a computer. AI painted the bird simply by text description, this is a great success in the development of such technologies.
Of course, while this is far from ideal. But we are seeing rapid progress in the development of various scientific achievements, including the development of AI. With ease you can imagine how these abilities of artificial intelligence in the future will help artists and designers. After all, visualization requires a minimum amount of time, and the result can exceed all expectations. This can also help in the processing of photos using voice control.
Recently, Japanese researchers began to teach AI visualization of human thoughts. This is achieved with the help of neural networks. While the images are blurred and more reminiscent of blots. But there is a direct relationship between the thoughts of man and this image. Such technology can be used in the future, for example, to study dreams. Also – for art, because the imagination of a person can be quickly realized. In addition, this is a way of reading thoughts – a serious specificity.
Microsoft generally assumes that it will be possible to create animated films only on the basis of a text script. It sounds, of course, almost fantastically, but soon there will be nothing to be surprised.
The team of researchers began developing a computer vision and natural language processing. These technologies can greatly help blind people. The AI creates captions for photos, and then, with the help of the received data, can answer questions posed to users about what is painted.
One aspect is very difficult for developers. Man, when perceiving visual images, uses such quality as attention. This is the team’s most attention and tries to simulate in AI. It is necessary to make attention computable. So that attention becomes math. The fact that AI painted a bird by description is already a big breakthrough in this area.
