Geo Ahn
Looking for Ph.D. positions starting in Fall 2027
I recently received my M.S. in Computer Science & Engineering from Kyung Hee University, South Korea, where I was advised by Prof. Jinwoo Choi. I was a research intern at NAVER Cloud (Video Team) in 2025, and continue to work with Inwoong Lee, Taeoh Kim and Minho Shim. I am also fortunate to collaborate with Prof. Joonseok Lee at Seoul National University.
My research centers on how to best represent videos so that models can perceive the real world as humans do. I have mainly addressed this in video understanding tasks with both unimodal and multimodal models. Recently, I have been working on latent action world models, asking whether their learned actions capture real-world dynamics.
Publications
* equal contribution
-
EVIDENT: Routing MLLM Adaptation through Entity-Grounded Visual Evidence for Cross-Domain Video Temporal GroundingIn NeurIPS, 2026An early version, SlotVTG, appeared in GRAIL-V Workshop at CVPR 2026 -
Why Can’t I Open My Drawer? Mitigating Object-Driven Shortcuts in Zero-Shot Compositional Action RecognitionIn ECCV, 2026 -
DEVIAS: Learning Disentangled Video Representations of Action and Scene for Holistic Video UnderstandingIn ECCV (Oral, 2.3% acceptance rate), 2024