LG unveils 'captioning' AI at computer vision conference

Lee Hong-lak, chief scientist of AI at LG AI Research, explains LG's Zero-shot Image Captioning technology during the Computer Vision and Pattern Recognition 2023 event in Vancouver, Sunday (local time). Courtesy of LG Corp.

By Baek Byung-yeul

LG unveiled Captioning AI, a generative AI service that recognizes elements in an image and generates descriptions and keywords, raising expectations about how the Korean conglomerate can impact the market where generative AI services such as ChatGPT are driving change, according to LG Corp., Monday.

LG Corp., a holding company of LG Group, said LG AI Research unveiled its Captioning AI service at the Computer Vision and Pattern Recognition 2023 event, the world's largest computer vision conference held in Vancouver, Canada, Sunday (local time).

LG said the service is based on LG AI Research's Zero-shot Image Captioning, a technology that enables AI to understand and describe objects or scenes it sees for the first time using its previous experience and knowledge, just as humans do.

The company explained that Captioning AI is different from AI services such as Midjourney, where users type in text or insert an image file and the AI draws a picture.

“Captioning may seem simple because it's an old concept, but the idea of applying generative AI technology to captioning is that AI has the visual intelligence to make inferences about images it hasn't seen before. For example, it can look at a landscape or a person in an image and deduce the location,” a spokesman of LG Corp. said.

The image shows LG's Zero-shot Image Captioning technology recognizing various elements and features in images to automatically generate descriptions and keywords. Courtesy of LG Corp.

Captioning AI can generate text descriptions and keywords for 10,000 images in less than two days, which can increase work efficiency and productivity for companies that need to manage large volumes of images, according to researchers.

The service was possible through collaboration with Shutterstock, the world's largest platform for visual content including images and videos. LG AI Research worked with the U.S.-based company, which has vast know-how in image capturing. The two sides also worked on securing copyright transparently and verifying AI ethics, such as whether AI collects data in a biased way when learning images and sensationalism issues.

“To secure global research leadership in image capturing, we plan to continue developing new metrics and researching new technologies by establishing organic collaborations with various partners,” said Kim Seung-hwan, leader of LG AI Research's Vision Lab.

During the conference period, LG Group affiliates such as LG AI Research, LG Electronics, LG Innotek, LG Energy Solution and LG Uplus hosted LG AI Day, a recruitment event for graduate students.