Tae-Hyun Oh

This work shows how to generate useful synthetic datasets for optical flow estimation.

Noise Map Guidance improves the quality of image editing by retaining spatial context.

Improving the way we identify sound sources using audio-visual data.

ObjectDR generates paired data to improve 3D shape reconstruction from 2D images.

New model enhances real-time video analysis with effective motion magnification.

A study on improving sound source localization by better using audio and visual information.

A new benchmark sheds light on hallucination in vision language models.

This article investigates how VLMs perceive color, shape, and meaning in images.