Stony Brook University Computer Science

An undergrad research project that began in the SUNY Korea CS Department has led to a paper accepted at the International Conference on Pattern Recognition, ICPR, a major international venue in pattern recognition and computer vision.

Yerin Cheon started the work as part of her CSE487 research project, co-supervised by Prof. François Rameau and Prof. Aruna Balasubramanian from Stony Brook University, who was visiting SUNY Korea at the time. After graduating from SUNY Korea, Yerin joined Stony Brook University for her master’s degree, where she continued developing the project in Prof. Aruna Balasubramanian’s group. The resulting paper, “Dual-Foundation Models for Unsupervised Domain Adaptation,” is the outcome of this continued collaboration.

The paper addresses a core challenge in robotics and autonomous vehicles: safely navigating an environment requires understanding the surrounding scene. This is often achieved through semantic segmentation, where each pixel in an image is assigned a category label. Training such deep learning models typically requires large amounts of pixel-level annotation, and the resulting models often do not transfer well to new environments. Unsupervised domain adaptation aims to mitigate that issue, but existing methods still struggle to produce reliable pseudo-labels and robustly align features across domains. Yerin’s approach tackles this limitation by leveraging vision foundation models to improve both pseudo-label refinement and feature alignment.

The work is a strong example of how undergraduate research at SUNY Korea can develop into peer-reviewed international publications and continue through collaboration with Stony Brook University.

Research Information
· Paper Title: Dual-Foundation Models for Unsupervised Domain Adaptation
· Conference: ICPR 2026
· Authors: Yerin Cheon, Aruna Balasubramanian, and Francois Rameau
· Project page: https://github.com/ycheon1101/DFUDA
· Paper: https://arxiv.org/pdf/2605.03365