This article is the NetEase smart studio (public number smartman163) produced. Focus on AI and read the next big era!
[NetEase smart news January 22 news] Microsoft recently introduced a new artificial intelligence technology that mimics similar artists - a "drawing robot." The robot can create an image corresponding to the text description, and it also adds details that go beyond itself and not just in the corresponding text. "These phenomena may indicate that artificial intelligence has its own imagination," said a staff member at Microsoft.
"If you search for information on a bird in the Bing engine, you will get a bird-related picture. But here, the picture was created by the computer from scratch pixel by pixel," Microsoft Corp. in Washington Ray He Xiaodong, head researcher and research manager of the Deep Learning Technology Center at the Microsoft Research Laboratory in Monde, said in a recent Microsoft announcement. "These birds may not exist in the real world, and they only represent one aspect of bird imagination that we developed."
Researchers say that robots can generate a variety of images, including from "normal idyllic scenes," such as grazing livestock, and even include imaginary diagrams of "floating double-decker buses."
According to Microsoft, the robot has trained on the data sets of the paired images and titles. This training enables them to understand how to match the corresponding words and images. For example, when the title says "bird", it first learns to draw a bird, and then uses machine learning to understand what the bird's image should look like.
He Xiaodong said: "This is one of the fundamental reasons why we believe machines can learn."
The technology of the drawing robot consists of two machine learning models. One is to generate an image from a text description, and the other is to use a text description to judge the authenticity of the generated image. The former tries to get virtual photos from the latter, but the latter does not want to be fooled. Therefore, by combining "interactions" with each other, the two can jointly create higher-quality images.
It is particularly good at drawing images from more complex sentences, while other techniques may draw a bird from a title marked "bird", for example, if you ask it to draw a green crown, yellow wings and red Belly, then the quality will drop. Before Microsoft developed the technology, the general result was the production of a vague "green-yellow reddish bird", explained Microsoft's staff.
Particularly interesting is how the robot fills the information gap when no specific details are mentioned. The fact is, basically, because it can memorize training data, it has its own common sense to develop imagination. In this example of searching for birds, even if it is not stated in the text, the robot usually draws a bird sitting on a branch, because the image originally given to it for learning and memory usually shows something similar.
According to a recent research report, Microsoft also pointed out that compared with the existing technology, the image quality generated by this new type of robot is nearly three times higher than before.
Of course, this is not the first artificial intelligence technology developed and combined with art.
The combination of the two sometimes produces excellent results. For example, the images generated by Google's artificial intelligence machine show its artistic expression potential. Google also has a neural network that can guess what you are drawing. They also have an automatic drawing robot and regularly describe in detail how it can help the machine to draw.
Facebook has also been developing neural networks to make small pictures of airplanes, cars and animals, and even used it to create its own Bitmoji image from photos.
For Microsoft, teaching a robot to draw corresponding images based on text means that it has reached the technology needed in this field of computer vision and natural language processing.
This includes CaptionBot's development of automatically writing photo captions, and techniques that can answer people's questions about the image, such as the location or attributes of objects in the picture, which is helpful for blind people.
As for how artificial intelligence artists serve people in the real world, Microsoft has some ideas.
This shows that the robot can be used as a sketch assistant for painters or interior designers, or it can be used as a tool for sound-controlling and beautifying photos. ("Cortana, please draw a bird for me", maybe it can do it?)
He Xiaodong said that with the increase in computing power, the technology may help film animation, thereby reducing the animator's manual labor required for film post-production.
But the technology has not yet reached the level where we can use it to complete the proposed requirements.
If you look closely at these images, they are almost always flawed. We can clearly see that they were created by machines, not humans: for example, blue ostriches, fruits, and grotesque bananas (above). .
However, with its 3 times faster speed capability, drawing robots represent a milestone in the development of artificial intelligence, Microsoft officials said.
(From: TechCrunch Compilation: NetEase Smart Participation: Fu Zeng)
Air Powered Stirrer,Vertical Air Mixer,Vertical Pneumatic Stirrer,Industrial Pneumatic Air Mixer
RUDONG HONGXIN MACHINERY CO.,LTD , https://www.rdhxmfr.com