UNIST Professor Baek Seung-ryeol's Team Wins Two Awards at the European Conference on Computer Vision (ECCV)

A master's student, Jo Hoseong, from the Graduate School of Artificial Intelligence at UNIST is taking a commemorative photo after participating in the HBHA competition at the European Conference on Computer Vision (ECCV) 2022.

[Asia Economy Yeongnam Reporting Headquarters Reporter Hwang Du-yeol] Professor Baek Seung-ryeol's team from the Graduate School of Artificial Intelligence at UNIST achieved excellent results at the HBHA competition hosted by the European Conference on Computer Vision (ECCV) 2022 on October 24.

The HBHA (Human Body, Hands and Activities from Egocentric and Multi-view Cameras) competition is a platform where models for related applications such as 3D body pose, hand pose, and action recognition are evaluated using first-person and multi-view cameras.

This competition consisted of four challenges, and Professor Baek Seung-ryeol's team won awards in two challenges: hand pose estimation and action prediction.

The first challenge involved predicting hand poses from a first-person perspective. Professor Baek's team presented a model that predicts hand poses from a single RGB image without additional post-processing, winning first place.

The second challenge was to predict actions from a first-person perspective. Professor Baek's team proposed a model that predicts actions using hand poses and object poses as clues, earning second place.

Kim Dong-wook, a master's course researcher at the Graduate School of Artificial Intelligence at UNIST, said, "Previously, the method involved selecting and improving the hand position and pose most similar to the hand in the video from predefined options, but the prediction model we proposed this time can predict more accurate hand poses without predefined assumptions."

Jo Ho-sung, a master's course researcher at the Graduate School of Artificial Intelligence at UNIST, explained, "In first-person action prediction, changes in hand poses and object poses are important clues, and using the model we proposed this time, actions can be predicted."

Professor Baek Seung-ryeol of the Graduate School of Artificial Intelligence said, "Our lab has presented hand pose and action recognition technologies recognized for competitiveness not only domestically but also globally," adding, "We will continue to devote ourselves to research and strive to consistently achieve good results."

ECCV is a world-renowned computer vision conference where researchers from around the globe gather to share research achievements and information and discuss the future of computer vision industry and technology. The conference began in 1990 and has held 17 annual events.

At this conference, about 6,000 papers were presented in fields such as 'motion and action recognition,' 'gestures and poses,' 'datasets and evaluation,' 'motion tracking,' and 'neural generative models.'

Dongwook Kim, Master's student at the Graduate School of Artificial Intelligence, UNIST (left), Seungryeol Baek, Professor at the Graduate School of Artificial Intelligence, UNIST