Jeong Jin-woo, a student on Professor Kim Jangho’s research team in the Department of Artificial Intelligence at Kookmin University, has developed “TAQ (Temporal-Aware Quantization),” a video super-resolution quantization technology that can be efficiently deployed in real-world systems, through collaborative research with students Ahn Sang-ho and Jeong Seong-yeop, as well as the LG CTO Lab. This research was presented at ECCV 2026 (European Conference on Computer Vision), a world-renowned international conference on computer vision.
 
Although video super-resolution models offer high performance, they require significant computational power and memory, limiting their application in mobile and edge environments. To address this, the research team proposed a new method that utilizes the temporal characteristics of video in the quantization process.
TAQ optimizes quantization parameters by utilizing temporal information from multiple frames during an offline calibration process. Notably, during actual inference, a single static quantization model can be used without the need to change quantization settings for each frame, offering the advantage of being easily deployable in existing hardware and AI accelerator environments.
 
This research is significant in that it considers not only quantization performance but also deployability in real-world systems. It is expected to be expanded in the future to include video restoration in mobile and edge environments, as well as lightweight technologies for various video-based AI models.
 
Meanwhile, this research was conducted through an industry-academia collaboration between Kookmin University and the LG CTO Lab, and the paper is titled “TAQ: Static-Deployable Temporal-Aware Quantization for Real-World Video Super-Resolution.”