서강대학교 전자공학과

Publishes a Paper at the Top Conference 'ICCV 2023' in the Field of Artificial Intelligence

▲ (from left) Professor Kang Suk-ju of the Department of Electronic Engineering, Yang Chang-hee in the master’s course, and Professor Kong Kyeong-bo of Pukyong National University (Sogang graduate)

A research team led by Professor Kang Suk-ju of the Department of Electronic Engineering (Yang Chang-hee in the master’s course and Professor Kong Kyeong-bo of Pukyong National University) conducted collaborative research with the NAVER Cloud team and published a paper at ICCV 2023, a top conference in the field of artificial intelligence.

ICCV (International Conference on Computer Vision), organized by IEEE/CVF, is one of the top conferences in the field of computer science and is a prestigious conference on computer vision and artificial intelligence-pattern recognition. ICCV 2023 will be hosted from October 2 to October 6 at the Paris Convention Center.

The team’s paper is titled SEFD: Learning to Distill Complex Poses and Occlusions, and the research was conducted in collaboration with NAVER Cloud.

In this paper, the research team proposed SMPL Edge Feature Distillation (SEFD) to resolve difficulties in measuring occlusion and complex poses. The results were shown to be efficient and competitive with other technologies. The paper was also highly recognized for its novel approach to the concept of existing knowledge distillation techniques and for its design which showed efficient results in the practical testing phase.

“I’m very honored that my paper is listed in the ICCV during my graduate studies, and I hope it will motivate other graduate students and researchers at Sogang University,” said Yang Chang-hee in the master’s course. “I hope that more of our students will submit papers to top-tier international conferences and achieve success in their work.”

▲ (a) How to create the SMPL edge map and how it works in the input stage

(b) How to distill fringe noise in the SMPL edge map

▲ Comparison photos of existing Stage-Of-The-Art (SOTA) and proposed methods

▶ Title of the paper: SEFD: Learning to Distill Complex Poses and Occlusions

▶ Author information: Yang Chang-hee (co-first author), Professor Kong Kyeong-bo (co-first author, Pukyong National University), Min Seong-jun (co-first author, Samsung Electronics), Cha Gun-ho (co-second author, NAVER Cloud), Jang Ho-deok (co-second author, NAVER Cloud), Wee Dong-yoon (co-second author, NAVER Cloud), Professor Kang Suk-ju (corresponding author, Sogang University)