The research team of Professor Kang Suk-ju, Department of Electronic Engineering,

Paper Accepted at 'WACV 2024', a Premier Conference in Computer Vision

▲ (Top row, from left) Professor Kang Suk-ju, Department of Electronic Engineering; Kang Byeong-woo, master's student

(Bottom row, from left) Moon Seung-hoon, combined master's–doctoral student; Cho Yu-bin, master's student; Yoo Hyun-woo, combined master's–doctoral student

A paper presented by the research team of Professor Kang Suk-ju of the Department of Electronic Engineering (master's student Kang Byeong-woo of Electronic Engineering, combined master's–doctoral student Moon Seung-hoon of Electronic Engineering, master's student Cho Yu-bin of the Department of Artificial Intelligence (DHE), and combined master's–doctoral student Yoo Hyun-woo of Electronic Engineering) has been finally accepted at WACV 2024, a premier conference in computer vision.

The paper, titled 'MetaSeg: MetaFormer-based Global Contexts-aware Network for Efficient Semantic Segmentation', proposes CRA (Channel Reduction Attention), a self-attention mechanism that greatly reduces computation by shrinking the feature channels of query and key while still outperforming existing methods.

The team further proposed MetaSeg, a semantic segmentation network that incorporates into the decoder a GMB (Global Meta Block) implementing CRA as a token mixer, achieving the best semantic segmentation performance at low computational cost.

▲ The CRA mechanism proposed in the paper

▲ Comparison of semantic segmentation performance against existing state-of-the-art (SOTA) methods

WACV 2024, organized by IEEE/CVF, will be held from 4 to 8 January 2024 at the Waikoloa Beach Marriott Resort, Hawaii.

▶ Paper title: MetaSeg: MetaFormer-based Global Contexts-aware Network for Efficient Semantic Segmentation

▶ Authors: Kang Byeong-woo (joint first author), Moon Seung-hoon (joint first author), Yoo Hyun-woo (joint first author), Cho Yu-bin (joint first author), Professor Kang Suk-ju (corresponding author, Sogang University)

Source: Sogang People https://www.sogang.ac.kr/research/res_01.html