[Edaily Reporter Shin Ha-yeon ] C-Lab ( XIIlab Co., Ltd(189330)), a company specializing in AI video analysis and digital twins, took first place on the public leaderboard in the text-based person search category of the global computer vision competition “AI City Challenge 2026.”

XIIlab Co., Ltd announced on the 15th that it secured first place on the public leaderboard among all 28 participating teams in Track 4 (text-based person anomaly detection, Sim2Real) of the AI City Challenge 2026, organized by NVIDIA, with a mean average precision (mAP) of 99.30%.

Now in its 10th year, the AI City Challenge is a global AI competition held in conjunction with the European Conference on Computer Vision (ECCV) 2026 workshop, where teams compete to develop AI technologies that solve real-world urban challenges such as intelligent transportation and smart cities. The XIIlab Co., Ltd. team “Xiilab.Alpex” achieved a 100% Recall@10—the proportion of correct individuals included in the top 10 search results—and a 98.74% Recall@1—the proportion of cases where the first search result was correct.

This track evaluates text-based person anomaly detection technology that identifies a specific person across multiple camera feeds using only natural language descriptions, such as “a person wearing a yellow shirt who has collapsed on the street.” The “Sim2Real (Simulation-to-Real)” approach was applied, requiring the AI—trained exclusively on approximately 1.01 million synthetic images—to accurately identify the target person in approximately 37,000 real-world images.

XIIlab Co., Ltd has been developing technology that uses NVIDIA Omniverse-based digital twins to recreate factories, logistics centers, and urban spaces in virtual environments, and then trains Vision AI using synthetic data generated within these environments. The company generates data on hazardous situations—such as falls or assaults—that are difficult to obtain in the real world within digital twin environments and utilizes it for AI training.

The competition is currently undergoing final rule verification procedures, and the final rankings and awards are scheduled to be announced on-site at ECCV 2026 on September 8–9.

XIIlab Co., Ltd plans to apply these results to its Vision AI products and digital twin business. The company aims to enhance its video analysis solutions—which can be applied in real-world environments—by recreating customer sites as digital twins and training AI using synthetic data.

Song Yu-jin, CTO of XIIlab Co., Ltd, stated, “The core of Sim2Real is whether AI trained in a virtual environment operates identically in the real world,” adding, “The fact that a model trained solely on synthetic data ranked first in the real-image evaluation proves that XIIlab Co., Ltd’s data generation technology is at a level suitable for field application.”