LG Group to boost image captioning technology

[Image source: LG AI Research]

LG Group is poised to engage deeper in artificial intelligence (AI) technology, namely image captioning that describes the content of an image in words, as it sees the area growing in the image search sector.

LG AI Research said on Tuesday that it will hold an “LG Global AI Challenge” event for global AI researchers until the end of April. The event, under the theme of “Zero-shot image captioning,” will be hosted with Seoul National University (SNU) Interdisciplinary Program in Artificial Intelligence and U.S. stock photography service Shutterstock.

AI researchers will compete to see how their AI programs handle the understanding of and explanation of an image shown to them for the first time. “Zero-shot image captioning” refers to a technology that enables AI programs understand and explain in written words an image that the programs have never “seen” before, based on a range of data that it previously had fed into it.

For example, when a person who sees a rabbit for the first time sees a rabbit and a cat together, the person may learn that there are shared traits and differences between the two mammals, and deduce something like, “Rabbits and cats both have fur, but only rabbits have long ears and their hind legs are more developed than cats.”

It is hoped that “zero-shot image captioning” can bring both accuracy and fairness to AI in understanding an image. It can allow an AI program to process immense amounts of online data and automatically create a caption or keywords for an image, from which humans will benefit concerning the convenience and accuracy of our online search results.

An AI program that has been fed professional medical data, for example, might be able to analyze medical images, as well.

Shutterstock, one of the largest stock image platforms, provided 26,000 datasets for the challenge. They consist of images and texts about the images and have passed AI morality tests. The images are of various forms, from photographs to illustrations and graphics.

LG Group expects image captioning technology will completely change the nature of image searches.

Its AI research lab is also hosting a workshop titled, “New Frontiers for Zero-Shot Image Captioning Evaluation,” at Computer Vision and Pattern Recognition (CVPR) conference that will take place in June in Vancouver, Canada. The CVPR conference is one of the world’s largest conferences in the field of computer vision. “Image captioning is one criterion that shows us how close AI is to human intelligence,” said Lee Kyoung-mu, a professor at SNU. “This challenge and workshop indicate that Korea’s AI capacity has reached a global level.”