简体中文
EnglishWhat Research Topics and Papers Are Featured on Rex Cheng's Website?
Updated
hkchengrex.com is the personal academic site of Ho Kei (Rex) Cheng, a research scientist at Luma AI working on multimodal foundation models. The site's "Selected Research" section lists his representative papers, each with links to a project page, code, and arXiv preprint where available. The work clusters around computer vision and generative modeling: segmentation and tracking in images and video, video-to-audio synthesis, and flow-based generation.
Research areas covered
The featured papers fall into three broad themes:
- Segmentation and tracking — unified models for detecting, segmenting, and tracking objects across images and video, using text, exemplar, and visual prompts.
- Video-to-audio synthesis — generating synchronized audio from video or text inputs.
- Generative modeling and optimal transport — improving conditional flow-based generation through better coupling of prior and data distributions.
Featured papers
| Paper | Venue | Topic |
|---|---|---|
| SAM 3: Segment Anything with Concepts | ICLR 2026 | Unified detection, segmentation, and tracking in images and video via text, exemplar, and visual prompts |
| The Curse of Conditions: Analyzing and Improving Optimal Transport for Conditional Flow-Based Generation | ICCV 2025 | Condition-aware coupling for straighter flows, avoiding test-time degradation from naïve optimal transport |
| MMAudio: Taming Multimodal Joint Training for High-Quality Video-to-Audio Synthesis | CVPR 2025 | Synchronized audio from video or text; trains on multi-source data even when some modalities are missing |
| Putting the Object Back into Video Object Segmentation | CVPR 2024 (Highlight) | Object transformer combining pixel-level and object-level features for efficient, robust video object segmentation |
Each entry links to a project page, code repository, and arXiv paper; MMAudio additionally links a Space demo and Replicate.
Background and affiliations
Cheng is currently a research scientist at Luma AI. He completed his PhD at the University of Illinois Urbana-Champaign, advised by Alexander Schwing, and interned at Adobe Research, Kaiber, Sony AI, and Meta FAIR/MSL. He was previously at The Hong Kong University of Science and Technology, advised by Yu-Wing Tai and Chi Keung Tang. The site links to his GitHub, Google Scholar, and CV.
How to use the site
If you want to follow a specific line of work, start from the paper entry and use its code or project page link rather than searching elsewhere — the site groups the canonical links for each project in one place. For current affiliation or publication updates, the CV and Google Scholar links are the most reliable starting points.