South Korean startup Mobiltech took first place in the “Building3D Challenge,” a digital twin competition organized by CVPR, the world’s most prestigious academic conference in the field of computer vision. The company outperformed teams from the Hong Kong University of Science and Technology (HKUST) and Xi’an Jiaotong University, both regarded as top global artificial intelligence (AI) institutions.

According to Mobiltech on the 12th, the Building3D Challenge is a competition that evaluates technologies for reconstructing cities in three dimensions (3D) using approximately 5 million building data points and AI. The key evaluation criterion is how precisely actual building shapes can be restored based on point cloud data collected with LiDAR sensors.

Mobiltech recorded an error about 61% lower than the second-place team in “Chamfer Distance,” the core metric representing building shape restoration accuracy. A lower Chamfer Distance value indicates that the restored result is more similar to the real building.

The second and third places in this competition went to the teams from HKUST and Xi’an Jiaotong University, respectively. Both universities are regarded as having world-class research competitiveness in computer vision and AI.

Mobiltech utilized its AI-based Automatic Feature Extraction (AFE) technology to automatically recognize various objects such as buildings, roads, and roofs from point cloud data and render them as 3D models. This technology is used as a core foundational technology for building digital twins for autonomous vehicles, robots, and physical AI development.

Kim Jae-seung, CEO of Mobiltech, said, “The quality of automatic 3D modeling is determined by how accurately objects are extracted,” adding, “This achievement demonstrates Mobiltech’s capabilities in a global research environment.”

Popular News