This AI system will create images from the input text

At the present time, artificial intelligence is proving that they are extremely good at two areas: making logical decisions and identifying patterns.

Thanks to machine learning - which helps AI teach itself based on selective input, we now have software that accurately predicts a person's personality just by their eye movement.

And yet, they can also write fiction, like the new episode of the magical boy Harry Potter and Portrait of the Giant Ash. Recently, a research team has successfully built a system that enables the conversion of text input into a sequence of images. The purpose of this project sounds very creative and groundbreaking, but the results are . a bit creepy.

This giraffe is green, pink and red, with a brown mark on it, and he likes to watch the sunset with his friends.

In fact, all the attempts received a rather peculiar response from AI, the animals in the picture all have a strange shape."A blindfolded cat and a blue mouth" , artificial intelligence produces a cat claw shaped by a cat's mouth.

This dolphin is glaring at his rival.

This AI system has been posted on the web thanks to New York University's Cristobal Valenzuela team who build free machine learning tools for the community. This software was actually built by Lehigh's local team of scientists. They hope to write down algorithms to improve the ability to identify and understand images of machine learning programs.

This kind of artificial intelligence (called AttnGAN ) has achieved quite impressive results. With a long sentence, such as "this bird is red and white, its mine is very short", it can produce a sequence of images corresponding to each part of the sentence.

The celery branch is burning on the fire.

Instead of just having a learning network to identify or export images, GAN uses 2 networks, one to create images, two to adjust it to match the description. This allows both to work smoother, smoother during learning, helping to speed up the process and create more accurate images.

As the author himself mentioned "automatically creating images based on natural language description is a fundamental problem" . They explained that the usual method for this type of AI is to evaluate the sentence or word file to try to find out what the required image is. The result of this method is okay, no.

Bananas and grapes are fighting.

Therefore, they decided to follow the new direction, which is an analysis of every part of the sentence being entered. This is an extremely complex method because basically, machines must mimic what the human mind will do in that situation thanks to a multitude of different algorithms.

This is not a perfect product, but it is a significant step forward. Thanks to the extensive knowledge of the whole school, this AI software is completely superior to other imaginative intelligence. The idea behind this project is that recognizing images is one thing, but making an image out of your own mind? That is another story.