LG to Enhance AI for Image Understanding and Explanation... Hosting Global Competition

Evaluating AI's Image Understanding Ability... Ongoing Until April

[Asia Economy Reporter Sunmi Park] LG is hosting the ‘LG Global AI Challenge’ for AI researchers worldwide. This initiative aims to lead the convergence trend of visual AI and language AI.

On the 31st, LG AI Research Institute announced that it will hold the ‘LG Global AI Challenge’ online from April 1 to the end of April. The competition focuses on ‘zero-shot image captioning,’ evaluating how accurately AI can understand and describe images it has never seen before. ‘Zero-shot image captioning’ is a technology where AI, similar to human visual cognition, can view images containing objects, animals, or landscapes it has never encountered and, based on learned data, independently understand and infer the content to describe it in text.

As ‘zero-shot image captioning’ technology advances, the accuracy and fairness of image recognition AI will improve, ultimately leading to the development of technologies that can directly assist people in their daily lives.

LG expects that zero-shot image captioning, which approaches human visual cognition, will significantly contribute to the technology development ecosystem of ‘EXAONE,’ a massive multimodal AI capable of expressing images as text and visualizing text as images. LG AI Research Institute is conducting this competition together with the ‘Seoul National University Graduate School of AI,’ which is researching the massive multimodal AI ‘EXAONE’ at a joint research center, and ‘Shutterstock,’ which is jointly preparing commercial services for image captioning AI.

Professor Kyungmoo Lee, Chair Professor at Seoul National University Graduate School of AI, explained, “Zero-shot image captioning is a very challenging problem and a recently emerging research field. Hosting the world’s first challenge and workshop jointly means that Korea’s AI capabilities have already reached a global level.”

Meanwhile, LG AI Research Institute will hold a workshop titled ‘Pioneers of Zero-Shot Image Captioning Evaluation’ at ‘CVPR 2023,’ the world’s most prestigious conference in computer vision, held in Vancouver, Canada, this June. Alongside global scholars and industry experts conducting AI research at companies such as Google and Microsoft, the workshop plans to engage in in-depth discussions on the direction and scalability of image captioning technology research and AI ethics issues.