Just a few words of description, Microsoft AI has drawn a beautiful picture

Not only does it serve effectively in image search for search engines, but in the future also promises to help users quickly sketch the images in human thoughts with just a few words. description - that's what researchers at Microsoft are aiming for.

Google can teach artificial intelligence (AI) how to create a doodle, but drawing a complete picture is complicated for AI. It is hard to imagine that AI can draw pictures through descriptive words such as: 'yellow bird stands out with black wings, soft fur and beautiful little beak'.

However, researchers at Microsoft have developed a technology based on that, which is capable of producing amazingly descriptive textual images.

The system cannot find a realistic image based on your thoughts but can draw images based on your description.

Painting by birds by Microsoft artificial intelligence.(Photo: Microsoft).

'If you access Bing and search for a bird image, you'll see a lot of pictures about it. But here, images are created entirely by computers, without human intervention. Those birds may not exist in the real world, it's just a product of your own imagination, ' said researcher Xiaodong He in a study.

Although this technology has not been fully developed, in the future, this technology will be a great help to architects or artists to help sketch as quickly as possible. Further, Mr. He said his research could create animated films based on descriptions with available scripts.

The group has begun to research the future of natural language processing computers with CaptionBot (subtitle robot system can read any content from a photograph in human language, along with emotion. of the person in the picture).

Researcher Xiaodong He.(Photo: Washington University).

On this basis, the system to answer questions from people about images will be created, called SeeingAI (application that can identify money, describe images . support blinds ).

Current technology consists of two parts: the part that produces the known images such as the Generative Adversarial Network (GAN) and the image quality evaluation department.

The 'robot artists' trained in pairs of images - annotations with images and emotions of different characters will be associated with specific descriptions. This research is based on the way people describe when imagining a picture in the head such as: Red field, sharp beak, yellow wing .