Hyunbin Oh, first-year integrated master's/doctoral student, Hyunwoo Ha, second-year master's student, Junseong Kim, third-year integrated student, Byeonggi Kwon, first-year doctoral student, and adviser Professor Tae-Hyun Oh of POSTECH's Department of Electrical Engineering received an Encouragement Award in the Outstanding Paper Awards at the 35th Workshop on Image Processing and Image Understanding (IPIU 2023), organized by the Korean Institute of Broadcast and Media Engineers, for “Real-Time Full-HD Video Motion Magnification Model.” This marks AMI Lab's third consecutive year of awards, four papers in total, following the 33rd and 34th workshops.
Video motion magnification captures and amplifies subtle movement that is difficult to see, making it perceptible and providing valuable information for structural safety assessment, medical monitoring, and other purposes. Professor Oh's “Learning-based Video Motion Magnification,” an oral presentation at ECCV 2018, made a breakthrough by successfully modeling conventional signal-processing-based motion magnification with deep learning.
The new paper identifies that this earlier model, still regarded as producing the best image quality in motion magnification, is too slow for real-time Full-HD applications. To address this, the authors introduce a two-stage framework for analyzing and simplifying models composed of diverse modules: learning-based identification/removal of unnecessary components, followed by hyperparameter search in the reduced space. The resulting model magnifies Full-HD video in real time, running three times faster with 4.5 times less computation than the baseline, without sacrificing image quality.
IPIU, now in its 35th year, is one of Korea's largest image processing and computer vision events alongside KCCV. This year it published 280 papers. Professor Oh concurrently leads a RIST–POSCO Open Lab group on video motion amplification and continues active research service, including Area Chair for ICCV 2023 and Associate Editor of The Visual Computer, impact factor 2.8.