LG Unveils First Generative AI Commercial Service: 'Captioning AI' Describing Images

LG AI Research Institute's First Commercial 'Generative AI' Service
Five Major Affiliates Participate in the World's Largest Computer Vision Conference CVPR

LG's first commercial generative AI service has been launched. It is a 'Captioning AI' that can describe even unfamiliar images in natural language like a human. It generates metadata such as sentences or keywords that can be used for image search. Simply put, the AI looks at a picture and writes its own description of the image.

On the 18th (local time), LG AI Research unveiled the captioning AI, which will bring changes to the image search market, at the world's largest computer vision conference, CVPR (Computer Vision and Pattern Recognition) 2023, held in Vancouver, Canada. The AI applies 'zero-shot image captioning' technology, enabling it to understand and describe objects or scenes it has never seen before by leveraging prior experience and knowledge. This technology allows the AI to recognize various elements and features of an image?such as background, people, and actions?based on a large amount of previously learned images and texts, and to explain their relationships.

Captioning AI that recognizes various elements and features of an image to generate descriptive texts and keywords.

Captioning AI can improve the work efficiency and productivity of companies that need to manage large volumes of images. On average, it generates 5 sentences and 10 keywords within 10 seconds. When scaled to 10,000 images, the task can be completed within two days, enabling the rapid establishment of customized image search and management systems.

The development of captioning AI was made possible through close collaboration between LG AI Research and Shutterstock. Shutterstock is the world's largest platform company with hundreds of thousands of new visual contents such as images and videos added daily, staffed with experts experienced in analyzing and processing content. LG AI Research worked together with Shutterstock, which possesses extensive know-how on image captioning, including suitable sentence length and expression methods for image classification and search, from data training to service development, enhancing the completeness of the technology.

On the 18th (local time), Lee Hong-rak from LG AI Research CSAI explained the zero-shot image captioning technology applied to captioning AI at the workshop.

On the same day, LG AI Research held a workshop on 'Zero-shot Image Captioning,' the core technology behind captioning AI, together with Seoul National University's AI Graduate School and Shutterstock. The workshop began with an opening speech by Lee Kyung-moo, Chair Professor at Seoul National University. It was attended by world-renowned experts in image captioning, including Lee Hong-rak, Chief AI Scientist (CSAI) at LG AI Research and professor at the University of Michigan; Cordelia Schmid, Research Director at the French National Institute for Computer Science and Google Research; Jack Hessel, researcher at the Allen Institute for AI; Hamid Palangi, Senior Researcher at Microsoft Research and professor at the University of Washington; and Anna Rohrbach, researcher at UC Berkeley.

Meanwhile, the workshop also hosted the awards ceremony for the 'LG Global AI Challenge' held earlier this year. The 'LG Global AI Challenge,' a competition evaluating AI models' image understanding capabilities, attracted a total of 142 research teams. Participants from Nanjing University of Science and Technology and KAIST, who won first and second place respectively, presented their research achievements at the workshop.

During CVPR 2023, which runs until the 22nd, LG AI Research will also work with major affiliates such as LG Electronics, LG Innotek, LG Energy Solution, and LG Uplus to secure top global AI talent. From the 20th for three days, AI researchers and recruitment officers from each LG affiliate will conduct demonstrations of their latest AI technologies and recruitment consultations at the LG integrated booth.

This content was produced with the assistance of AI translation services.