# Chunhua Shen
**Chunhua Shen** (沈春华) is a computer vision researcher who works on semantic segmentation (Labeling every pixel of an image with its object class), object detection, and machine learning.[1](https://cshen.github.io/)
| Fact | Detail |
|---|---|
| Field | Computer vision and machine learning; semantic segmentation and object detection[3](https://person.zju.edu.cn/en/chunhua) |
| Training | BSc and MSc, Nanjing University (1995–2002); PhD in computer vision, University of Adelaide, 2006, supervised by Michael Brooks; MPhil in applied statistics, Australian National University, 2009[4](https://users.cecs.anu.edu.au/~cs/web/resume.html) • [5](https://www.linkedin.com/in/chunhua-shen-1b88711a6) |
| Fellowship | ARC Future Fellowship, 2012–2016[7](http://aia.hust.edu.cn/info/1180/6582.htm) |
| Current research | Unified open-world segmentation, multimodal large language models, monocular geometric foundation models[8](https://arxiv.org/pdf/2510.10524) • [2](https://cshen.github.io/paper.html) |
## Education and early career
From 1995 to 2002 Shen studied mathematics and physics (BSc) and speech signal processing (MSc) at Nanjing University in China.[4](https://users.cecs.anu.edu.au/~cs/web/resume.html) In 2006 he completed a PhD in computer vision at the [University of Adelaide](https://www.edgepedia.org/university-of-adelaide), supervised by Michael Brooks.[4](https://users.cecs.anu.edu.au/~cs/web/resume.html) • [5](https://www.linkedin.com/in/chunhua-shen-1b88711a6) In 2009 he added an MPhil in applied statistics from the [Australian National University](https://www.edgepedia.org/australian-national-university).[4](https://users.cecs.anu.edu.au/~cs/web/resume.html)
After his doctorate he joined the computer vision program at National ICT Australia (NICTA) in Canberra, where he was a senior researcher from October 2005 to March 2011 working on computer vision and statistical machine learning, including boosting methods for real-time object detection.[4](https://users.cecs.anu.edu.au/~cs/web/resume.html) • [5](https://www.linkedin.com/in/chunhua-shen-1b88711a6) • [6](http://aia.hust.edu.cn/info/1180/6290.htm) During those years he also held an adjunct position at the Australian National University's College of Engineering and Computer Science, from 2006 to 2011.[7](http://aia.hust.edu.cn/info/1180/6582.htm)
## Career in Australia
Shen moved back to the University of Adelaide in 2011 as a professor in the School of Computer Science, where he later served as Director of Research.[7](http://aia.hust.edu.cn/info/1180/6582.htm) In 2012 he was awarded an [Australian Research Council](https://www.edgepedia.org/australian-research-council) (ARC) Future Fellowship, which he held until 2016.[1](https://cshen.github.io/) • [7](http://aia.hust.edu.cn/info/1180/6582.htm) From 2014 he was a Project Leader for machine learning in robot vision and a Chief Investigator at the ARC Centre of Excellence for Robotic Vision, a national center supported by 20 million Australian dollars in federal funding over 2014–2020.[6](http://aia.hust.edu.cn/info/1180/6290.htm)
## Representative work
In semantic segmentation, *RefineNet*, a multi-path refinement network for dense prediction, was published in TPAMI in 2019, and *Exploring context with deep structured models for semantic segmentation* was published in TPAMI in 2017.[2](https://cshen.github.io/paper.html) In medical imaging, two IEEE Transactions on Medical Imaging papers addressed automated skin lesion analysis. The 2020 mutual bootstrapping model (MB-DCNN) performed segmentation and classification simultaneously through a coarse segmentation network, a mask-guided classification network, and an enhanced segmentation network; on the ISIC-2017 and PH2 datasets it reached Jaccard indices of 80.4% and 89.4% for segmentation and average AUCs of 93.8% and 97.7% for classification, and bootstrapping with the coarse network raised the classification AUC on the ISIC-2017 validation set from 94.9% to 97.0%.[10](https://doi.org/10.48550/arxiv.1903.03313) The 2019 companion paper introduced attention residual learning for skin lesion classification.[2](https://cshen.github.io/paper.html)
## Zhejiang University and current roles
According to his [Zhejiang University](https://www.edgepedia.org/zhejiang-university) personal homepage, Shen joined Zhejiang University as a Chair Professor in 2022 and is a doctoral supervisor in the College of Computer Science and Technology, with research listed in computer vision and machine learning.[1](https://cshen.github.io/) • [3](https://person.zju.edu.cn/en/chunhua)
## Research since 2024
His recent work has moved toward foundation models and multimodal systems. Metric3D v2, a monocular geometric foundation model for zero-shot metric depth and surface normal estimation, appeared in TPAMI in 2024, and a 2025 [International Journal of Computer Vision](https://www.edgepedia.org/international-journal-of-computer-vision) paper studied segment anything in context with vision foundation models.[2](https://cshen.github.io/paper.html) At CVPR 2025, SegAgent introduced the Human-Like Mask Annotation Task, in which multimodal large language models imitate human annotators using interactive segmentation tools framed as a multi-step decision process; on the ThinObject5K benchmark the StaR policy-improvement procedure raised IoU from a baseline of 71.45 to 86.57.[11](https://openaccess.thecvf.com/content/CVPR2025/papers/Zhu_SegAgent_Exploring_Pixel_Understanding_Capabilities_in_MLLMs_by_Imitating_Human_CVPR_2025_paper.pdf) In October 2025 his group released COSINE, a unified open-world segmentation model that consolidates open-vocabulary and in-context segmentation with text and image prompts, covering semantic and panoptic segmentation, referring segmentation, few-shot segmentation, and video object segmentation.[8](https://arxiv.org/pdf/2510.10524) His arXiv listing shows continued output into 2026, including ICCV 2025 papers and 2026 work on spatial intelligence.[2](https://cshen.github.io/paper.html) • [9](https://openreview.net/profile?id=%7EChunhua_Shen2)
## References
1. Chunhua Shen, personal homepage. https://cshen.github.io/
2. Chunhua Shen, Selected Publications. https://cshen.github.io/paper.html
3. Chunhua Shen, Zhejiang University faculty page. https://person.zju.edu.cn/en/chunhua
4. Chunhua Shen, ANU CECS resume page. https://users.cecs.anu.edu.au/~cs/web/resume.html
5. Chunhua Shen, LinkedIn profile. https://www.linkedin.com/in/chunhua-shen-1b88711a6
6. 学术报告201609期, HUST AI Institute seminar announcement, 2016. http://aia.hust.edu.cn/info/1180/6290.htm
7. 学术报告第201869期, HUST lecture announcement, 2018. http://aia.hust.edu.cn/info/1180/6582.htm
8. Unified Open-World Segmentation with Multi-Modal Prompts (COSINE), arXiv:2510.10524. https://arxiv.org/pdf/2510.10524
9. Chunhua Shen, OpenReview profile. https://openreview.net/profile?id=%7EChunhua_Shen2
10. A Mutual Bootstrapping Model for Automated Skin Lesion Segmentation and Classification, arXiv preprint of the IEEE TMI paper. https://doi.org/10.48550/arxiv.1903.03313
11. SegAgent: Exploring Pixel Understanding Capabilities in MLLMs by Imitating Human Annotator Trajectories, CVPR 2025. https://openaccess.thecvf.com/content/CVPR2025/papers/Zhu_SegAgent_Exploring_Pixel_Understanding_Capabilities_in_MLLMs_by_Imitating_Human_CVPR_2025_paper.pdf
---
*Topic: Encyclopedia › Technology and the built world › Engineers and computer scientists › Computer scientists and AI researchers › Researchers in artificial intelligence and machine learning › Computer Vision*
*Initially written Sep 21, 2026 · Reviewed: Sep 25, 2026 · Edited: Sep 25, 2026; Oct 15, 2026 · Last review: Sep 25, 2026*
*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*
License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license