简体中文
EnglishWhere to Find Rex Cheng's Code, Papers, and Academic Profiles
Updated
Rex Cheng's personal site, hkchengrex.com, is the central hub for his research output. From the homepage you can reach his GitHub, Google Scholar, and CV directly, and every listed paper links out to a project page, code repository, and arXiv preprint. If you want his papers, open-source implementations, or academic record, start there rather than searching each platform separately.
Who Rex Cheng is
Ho Kei (Rex) Cheng is a research scientist at Luma AI, working on multimodal foundation models. He completed his PhD at the University of Illinois Urbana-Champaign, advised by Alexander Schwing. During his PhD he interned at Adobe Research, Kaiber, Sony AI, and Meta FAIR/MSL. Before that he studied at The Hong Kong University of Science and Technology, advised by Yu-Wing Tai and Chi Keung Tang.
Links available from the homepage
The site's top-level navigation points to three profiles:
- GitHub — source code for his projects
- Google Scholar — publication list and citation record
- CV — full academic and industry history
These are the fastest routes to his code and publication record. The research entries below them repeat the same destinations at the paper level.
Research entries and where each one leads
Each selected research item on the page carries its own set of links. The pattern is consistent: a project page, a code repository, and an arXiv preprint, with some entries adding a demo.
| Paper | Venue | Links listed | What it covers |
|---|---|---|---|
| SAM 3: Segment Anything with Concepts | ICLR 2026 | Project page, code, arXiv | Unified model for detection, segmentation, and tracking in images and video using text, exemplar, and visual prompts |
| The Curse of Conditions: Analyzing and Improving Optimal Transport for Conditional Flow-Based Generation | ICCV 2025 | Project page, code, arXiv | Straighter flows via condition-aware coupling of prior and data samples, without the test-time degradation of naïve optimal transport |
| MMAudio: Taming Multimodal Joint Training for High-Quality Video-to-Audio Synthesis | CVPR 2025 | Project page, code, arXiv, Space demo, Replicate | High-quality synchronized audio from video or text, trainable on multi-source data with missing modalities |
| Putting the Object Back into Video Object Segmentation | CVPR 2024 (Highlight) | Project page, code, arXiv | Object transformer combining pixel-level and object-level features for efficient, robust video object segmentation |
How to use these links
- For the paper text: follow the arXiv link on the specific entry. That gives you the preprint version.
- For the implementation: follow the code link, which points to the GitHub repository for that project.
- For a quick trial without setup: MMAudio additionally lists a Space demo and a Replicate endpoint, so you can run it without cloning anything.
- For citation counts and the full publication list: use the Google Scholar link at the top of the page.
- For background and affiliation history: use the CV link.
One practical note: the page groups links per paper rather than in a single index, so if you are looking for a specific project, scan the research section for its title first, then take the link you need from that row.