Researchers at the Mobility Platform Research Center of the Korea Electronics and Technology Institute (KETI) won first place in the field of AI-generated image detection technology at an international competition held in conjunction with CVPR 2026, the world's largest computer vision conference. This achievement is noteworthy for simultaneously demonstrating generative AI content discrimination capabilities and inference performance in low-power environments.

KETI announced that it achieved the top ranking in the 'AI Generated Images Detection' category of the IEEE Low Power Computer Vision Challenge (LPCVC), held during the ECV (Efficient Computer Vision Deep Learning) workshop at CVPR 2026 in Denver, USA, from the 3rd to the 7th. CVPR is considered a leading international academic conference in the fields of computer vision and pattern recognition.

This competition evaluated not only the accuracy of distinguishing between images generated by generative AI and real images, but also the execution efficiency in edge AI environments with limited power and computational resources. In addition, whether the model can explain the basis of its derived results and whether it performs reasonable inference in various situations were included as major evaluation criteria.

The research team developed the technology based on a Vision-Language Model (VLM) that understands both images and text. A VLM is an artificial intelligence model capable of analyzing image content to explain it in natural language or provide the basis for judgments. The researchers performed lightweighting and optimization to enable the model to run on low-power edge devices while maintaining its performance.

In particular, it is reported that the company received high marks for securing explainability and inference capabilities, along with the accuracy of AI-generated image detection. Low-power inference technology is a technology that enables artificial intelligence to make decisions quickly and efficiently, even on devices using limited battery and computing resources.

Senior Researchers Jaewoong Yoo, Jinman Park, and Ganjerik from the KETI Mobility Platform Research Center participated in the award. The research team was responsible for model lightweighting, inference optimization, and designing a structure capable of providing the basis for judgment.

As the use of generative AI spreads, the demand for technology to identify fake images and forged or altered content is also growing. Researchers expect that this technology can be utilized in various fields in the future, such as autonomous driving, robotics, and industrial edge systems.

"This achievement is the result of confirming the competitiveness of the VLM-based recognition and judgment technologies and low-power inference technologies that the research center has accumulated on the international stage," said Park Bu-sik, Head of the Mobility Platform Research Center at KETI. "We will continue our research to continuously advance explainable edge AI technology so that it can be applied to various industrial fields."