On the 18th (local time), LG AI Research Institute unveiled ‘Captioning AI’, which describes images, for the primary time on the world’s largest computer vision conference ‘CVPR (Computer Vision and Pattern Recognition) 2023’ in Vancouver, Canada.

Captioning AI, a generative artificial intelligence (AI) commercialization service released for the primary time to the skin world, is ‘AI that may explain even the primary images seen like humans in natural language’, and meta data akin to sentences and keywords, that are information that could be used for image search. create

LG AI Research Institute applied ‘Zero-shot Image Captioning’ technology in order that AI can understand and explain objects or scenes that it sees for the primary time like humans using previous experience and knowledge.

Zero-shot image captioning is a technology that permits AI to acknowledge various elements and characteristics in images, akin to backgrounds, people, and actions, and understand and explain their relationships based on large amounts of previously learned image and text data.

It’s explained that Captioning AI can increase work efficiency and productivity for corporations that should manage large amounts of images. It will depend on the length and variety of sentences or words, but on average it generates 5 sentences and 10 keywords in 10 seconds. If the range of images is expanded to 10,000, the work could be accomplished inside two days, making it possible to construct a customized image search and management system in a brief time frame.

Captioning AI explained that it’s the results of collaboration with Shutterstock.

Shutterstock is the world’s largest platform company that adds tons of of hundreds of recent visual content akin to images and videos on daily basis, and has experienced experts in analyzing and processing content.

The LG AI Research Institute worked with Shutterstock, which has vast know-how on image captioning, akin to the length of sentences suitable to be used in image classification and search and learn how to express them, from data learning to service development to enhance the completeness.

Specifically, so as to develop a practical and reliable AI model, AI ethics verification akin to bias and selectivity of learning data was conducted, and copyright transparency was also secured.

Sezal Amin, CTO of Shutterstock, said, “Currently, we’re developing Captioning AI technology by conducting an ‘Early Access Program’ for 10 global customers. It can be an AI that helps you deal with your work.”

The LG AI Research Institute plans to display captioning AI services for researchers visiting the LG booth throughout the conference.

On today, LG AI Research Institute also held a workshop with Seoul National University AI Graduate School and Shutterstock with reference to zero-shot image captioning, the inspiration technology of captioning AI.

The workshop, which began with a gap speech by Seoul National University chair professor Kyung-Moo Lee, was attended by LG AI Research Institute’s top AI scientist, Professor Hong-Rak Lee on the University of Michigan, Cordelia Schmid, research director on the French National Institute of Computer Science and researcher at Google Research, Jack Hessel Allen, researcher on the Allen Institute for Artificial Intelligence, and Hamid Pallan. World-renowned experts in the sector of image captioning, akin to Microsoft Research Senior Researcher and Professor at Washington University and Ana Rohrbach, UC Berkeley Researcher, participated in an in-depth discussion on the most recent research trends, future prospects, and the impact of technology, akin to AI ethics, on society. proceeded

Meanwhile, at this workshop, the ‘LG Global AI Challenge’ award ceremony, which was held in the primary half of the yr, was also held.

A complete of 142 research teams participated within the ‘LG Global AI Challenge’, a contest that evaluates the image understanding ability of AI models developed in-house.

Participants from Nanjing University of Science and Technology and KAIST, who placed first and second within the challenge, also presented their research results on the workshop.

“This workshop is more meaningful since it is linked to the announcement of ‘captioning AI,’ the primary commercialized service,” said Kim Seung-hwan, head of LG AI Research Institute’s Vision Lab. We plan to proceed developing recent evaluation indicators and research on recent technologies by establishing a cooperative system.”

As well as, throughout the conference, which runs until the twenty second, the LG AI Research Institute will work with major affiliates of LG, akin to LG Electronics, LG Innotek, LG Energy Solutions, and LG U+, to secure outstanding global AI talents.

To this end, on the nineteenth, ‘LG AI Day’, a networking event for master’s and doctoral students who participated within the conference, was held, and from the twentieth to three days, AI researchers and recruiters from each LG affiliate held a The most recent AI technology demonstrations and recruitment consultations are held.

On the LG Integrated Booth, LG Electronics will use the motive force monitoring system, which is a technology that detects drowsiness and carelessness by recognizing the motive force’s face and gaze based on vision inspection technology, changes within the freshness of food within the refrigerator or changes within the condition of food within the oven in line with the cooking process. Introducing AI technology that implements visually.

As well as, LG Innotek developed digital twin technology that permits users to check products prematurely in a digital space before mass-producing actual products, and LG Energy Solutions developed Anomaly Detection, a vision-based inspection technology that detects defects in battery cells produced in tons of of hundreds a day. ), LG U+ will introduce AI technology that extracts metadata that expresses various information akin to people, actions, places, situations, and characters in video scenes in order that customers can easily find the scene they need in media content.

Reporter Jang Se-min semim99@aitimes.com