Mentors: Dr. Peimeng Guan & Dr. Dae Yeol Lee Manager: Dr. Guan-Ming Su

Greenfield research internship: proposed, scoped, and led my own research project.

• Proposed authoring text and icon annotations as native Gaussian assets inside volumetric video, so that inserted content shares the host scene’s rendering path and behaves correctly under depth ordering, mutual occlusion, and physically based relighting.
• Built the authoring system around a -based interactive scene editor, with a keyframe timeline and text tracks for timed, reusable annotations.
• Designed a feed-forward amortized authoring method in which a shared encoder–decoder predicts renderable TVGS assets directly from canonical anchor point clouds, keeping normal map prediction accuracy under 2 degrees while compressing the per-asset payload to a compact anchor group (roughly 28x smaller).
• Scaled training to a library of 27K+ physically based assets on multi-GPU infrastructure, generalizing to fonts never seen in training.

Mentor: Dr. Sri Harsha Musunuri Manager: Dr. Guan-Ming Su

Responsibilities include:

• Developed a geometry-guided framework to augment Video Large Language Models (VideoLLMs) with cinematographic camera motion understanding, resulting in a first-author publication accepted at CVPR 2026 Workshop PVUW.
• Created a synthetic video benchmark using Unreal Engine 5 and designed probing experiments to evaluate spatial retention in vision encoders via Q-former.
• Proposed a plug-and-play strategy that injects camera motion, framing, and spatial descriptors into VideoLLMs, improving reasoning for video captioning, VQA, and stylistic plagiarism detection.

Responsibilities include:

• Implemented LaTeX file compilation, encoding, and distribution features in the Moodle course management system as part of a copyright protection scheme.
• Developed web app user interfaces and system plugins for diverse applications using PHP and MySQL.
• Technical support for Moodle users, including troubleshooting and resolving software issues.