Hello, I'm Joseph Tung!

I am an CS Ph.D. student at NYU Courant, advised by Prof. David Fouhey. I received my Bachelor’s in Computer Science at Cornell University, advised by Prof. Noah Snavely. My interests are in 3D computer vision, specifically scene-level reconstruction and generation.

Email Scholar Github Twitter

News

April 2025: The DynPose-100K dataset is now available for download!
May 2024: I will be continuing my research this summer at Cornell Tech with Prof. Noah Snavely!

Publications

Emergent Extreme-View Geometry in 3D Foundation Models

Yiwen Zhang, Joseph Tung, Ruojin Cai, David Fouhey, Hadar Averbuch-Elor

We create a lightweight fine-tuning method for 3D foundation models to improve extreme-view geometry estimation, and release benchmarks for hard unconstrained image collections.

Project Page PDF arXiv

Dynamic Camera Poses and Where to Find Them

Chris Rockwell, Joseph Tung, Tsung-Yi Lin, Ming-Yu Liu, David Fouhey, Chen-Hsuan Lin

CVPR 2025

We create a large-scale, high-quality dataset of dynamic camera poses from 100K internet videos.

Project Page PDF arXiv Dataset

MegaScenes: Scene-Level View Synthesis at Scale

Joseph Tung*, Gene Chou*, Ruojin Cai, Guandao Yang, Kai Zhang, Gordon Wetzstein, Bharath Harharan, Noah Snavely

ECCV 2024

We create a dataset of 100K SfM reconstructions from 2M internet photos around the world. We use it to train a model for scene-level novel-view synthesis.

Project Page PDF arXiv Code Dataset Web-Viewer

Doppelgangers: Learning to Disambiguate Images of Similar Structures

Ruojin Cai, Joseph Tung, Qianqian Wang, Hadar Averbuch-Elor, Bharath Harharan, Noah Snavely

ICCV 2023 (Oral)

We train a classifier to disambiguate images that depict distinct, but visually similar structures, which we coin as "doppelgangers". We use this classifier to improve reconstruction quality in structure-from-motion.

Project Page PDF arXiv Code