彭芳 (Fang Peng)

华南农业大学
数学与信息学院、软件学院
首聘副教授
pengfang21@mails.ucas.ac.cn

彭芳，现任华南农业大学数学与信息学院、软件学院首聘副教授（青年引进人才），硕士生导师，主要从事人工智能、多模态大模型及计算机视觉相关前沿研究。

2026年于中国科学院自动化研究所获工学博士学位（模式识别与智能系统专业），师从徐常胜研究员（国家杰青、万人计划领军人才、IEEE Fellow），并由杨小汕研究员（国家优青）联合指导。此前于中国农业大学获学士与硕士学位，并曾在鹏城国家实验室从事大语言模型研发工作。

近年来，在 IEEE TMM、IEEE TCSVT、NeurIPS、ACM MM 等中科院一区 TOP 及 CCF-A 顶级期刊与会议上发表论文 10 余篇；其中单篇最高 Google Scholar 引用超 300 次，4 篇论文单篇引用超 100 次，总引用超 1200 次。主持国家自然科学基金青年项目C类 1 项、参与国家自然科学基金联合基金项目 1 项，并担任 IEEE TMM、ACM MM、CVIU、AAAI 等国际权威期刊和会议审稿人。

主要研究领域围绕多模态智能展开，重点关注：

1. 多模态大模型的后训练与推理增强

2. 大模型的轻量化与高效适配

3. 零样本 / 小样本学习

4. AIGC 驱动的开放世界视觉感知

5. 多模态智能体 (Multimodal Agents)

6. 视觉-语言-动作（VLA）模型

7. 智慧农业、智慧养殖（AI for Agriculture）

热忱欢迎计算机、数学、电子信息相关专业对人工智能感兴趣的本科生（科研实习/毕设）及推免、考研同学联系加入！无论是否有过科研经历，只要踏实勤奋、有自驱力、对前沿技术充满探索欲，均非常欢迎。课题组提供体系化的科研训练、充足算力及 OpenAI 前沿大模型的 Token 支持，共同探索多模态智能前沿。完成科研要求后鼓励学生开展实习；表现优异者可推荐至中国科学院自动化研究所（CASIA）等顶尖科研机构实习访问或深造。

诚挚欢迎相关领域的学者与同行开展交流和合作！

研究方向 (Research Interests)

多模态大模型的后训练与推理增强

大模型的轻量化与高效适配

AIGC驱动的开放世界视觉感知

教育背景 (Education)

中国科学院自动化研究所模式识别与智能系统 · 博士

2021.09 — 2026.01

多模态人工智能系统全国重点实验室（MAIS） · 导师：徐常胜研究员（国家杰青，万人计划领军人才，IEEE Fellow）

中国农业大学计算机科学与技术 · 硕士

2017.09 — 2020.01

信息与电气工程学院 · 导师：李振波教授

中国农业大学能源与动力工程 & 计算机科学与技术（双学位）· 学士

2013.09 — 2017.06

工作经历 (Experience)

华南农业大学 · 数学与信息学院、软件学院首聘副教授

2026.09 — 至今

鹏城国家实验室 · 开源软硬件平台

2020.04 — 2021.08

科研项目 (Research Projects)

国家自然科学基金青年项目C类：“面向开放域视觉理解的多模态大模型零样本推理方法研究”，2027.01 — 2029.12，主持。
国家自然科学基金联合基金项目：“面向多模态大模型训练的联邦学习算法”，2026.01 — 2029.12，参与。

代表性论文 (Selected Publications) 查看全部成果与架构预览图 →

Fang Peng, Xiaoshan Yang, Linhui Xiao, Yaowei Wang, Changsheng Xu*. SgVA-CLIP: Semantic-Guided Visual Adapting of Vision-Language Models for Few-Shot Image Classification. IEEE Transactions on Multimedia (IEEE TMM), 26: 3469-3480, 2024. CCF-A · 中科院一区 TOP IF = 9.7 引用 150+
Fang Peng, Xiaoshan Yang, Yaowei Wang, Changsheng Xu*. Progressive Learning of Instance-Level Proxy Semantics for Few-Shot Action Recognition. IEEE Transactions on Multimedia (IEEE TMM), 28: 853-864, 2026. CCF-A · 中科院一区 TOP IF = 9.7
Fang Peng, Xiaoshan Yang*, Yaowei Wang, Changsheng Xu. Group-Relative Visual Discrimination Enhancement for Unlocking Intrinsic Capability of MLLMs. IEEE Transactions on Circuits and Systems for Video Technology (IEEE TCSVT), 36: 5865-5876, 2026. 中科院一区 TOP IF = 11.1
Fang Peng, Zheng Miao, Fei Li, Zhenbo Li*. S-FPN: A shortcut feature pyramid network for sea cucumber detection in underwater images. Expert Systems with Applications (ESWA), 182: 115306, 2021. 中科院一区 TOP IF = 7.5 引用 100+
Linhui Xiao, Xiaoshan Yang, Fang Peng, Yaowei Wang, Changsheng Xu*. OneRef: Unified One-tower Expression Grounding and Segmentation with Mask Referring Modeling. Annual Conference on Neural Information Processing Systems (NeurIPS), 2024. CCF-A 类顶会
Linhui Xiao, Xiaoshan Yang, Fang Peng, Ming Yan, Yaowei Wang, Changsheng Xu*. CLIP-VG: Self-paced Curriculum Adapting of CLIP for Visual Grounding. IEEE Transactions on Multimedia (IEEE TMM), 26: 4334-4347, 2024. CCF-A · 中科院一区 TOP IF = 9.7 引用 100+

专利 (Patents)

授权发明专利：“小样本动作识别方法、电子设备及介质”，专利号：ZL 2025 1 0067424.8。
授权发明专利：“物体定位方法、装置、设备及介质”，专利号：ZL 2024 1 0382411.5。
授权发明专利：“物体检测分割方案”，专利号：ZL 2024 1 0272880.1。

Fang Peng, Ph.D.

Associate Professor · South China Agricultural University

Fang Peng

Ph.D. · Associate Professor

South China Agricultural University
School of Math & Information / Software
Email: pengfang21@mails.ucas.ac.cn
Guangzhou, China

Fang Peng is currently an Associate Professor (Young Talent Program) and Master's Supervisor at the School of Mathematics and Information / Software Engineering, South China Agricultural University. Her research focuses on artificial intelligence, multimodal foundation models, and computer vision.

She received her Ph.D. degree in Pattern Recognition and Intelligent Systems from the Institute of Automation, Chinese Academy of Sciences (CASIA) in 2026, advised by Prof. Changsheng Xu (National Distinguished Young Scholar, IEEE Fellow) and co-advised by Prof. Xiaoshan Yang. Previously, she received her B.E. and M.E. degrees from China Agricultural University, and conducted research on large language models at Peng Cheng Laboratory.

Her research centers on multimodal intelligence, with particular emphasis on post-training & reasoning enhancement of MLLMs, lightweight adaptation & parameter-efficient fine-tuning (PEFT), zero/few-shot learning, AIGC-driven open-world visual perception, multimodal agents, vision-language-action (VLA) models, and domain-specific foundation models for smart agriculture. She has published over 10 papers in prestigious journals and conferences such as IEEE TMM, IEEE TCSVT, NeurIPS, and ACM MM. Her Google Scholar citations exceed 1,200 (1 paper with >300 citations, 4 papers with >100 citations). She also serves as a reviewer for TMM, ACM MM, CVIU, AAAI, and is the Principal Investigator of an NSFC Youth Science Fund Project (Category C) as well as a participant in an NSFC Joint Fund Project.

Opening for Students: Highly motivated undergraduate and graduate students with a strong interest in multimodal AI, computer vision, and foundation models are warmly welcome to contact and join our group! We provide systematic scientific training, ample GPU computing clusters, and cutting-edge foundation model API tokens. Outstanding students can be recommended to top-tier research institutes (e.g., CASIA) for joint training or internships.

Collaborations and academic exchanges with colleagues worldwide are sincerely welcomed!

Research Interests

Post-training & Reasoning of MLLMs

Lightweight Adaptation & Efficient Tuning

AIGC-Driven Open-World Perception

Education

Institute of Automation, Chinese Academy of Sciences (CASIA) Ph.D. in Pattern Recognition & Intelligent Systems

2021.09 — 2026.01

State Key Laboratory of Multimodal Artificial Intelligence Systems (MAIS) · Advisor: Prof. Changsheng Xu (IEEE Fellow)

China Agricultural University M.E. in Computer Science and Technology

2017.09 — 2020.01

College of Information and Electrical Engineering · Advisor: Prof. Zhenbo Li

China Agricultural University B.E. in Energy & Power Engineering & Computer Science (Dual Degree)

2013.09 — 2017.06

Professional Experience

South China Agricultural University · School of Mathematics & Information / Software Associate Professor

2026.09 — Present

Peng Cheng Laboratory · Open Source Software & Hardware Platform Algorithm Engineer

2020.04 — 2021.08

Research Projects

National Natural Science Foundation of China (NSFC) Youth Science Fund Project (Category C): "Research on Zero-Shot Reasoning Methods for Multimodal Foundation Models in Open-Domain Visual Understanding", 2027.01 — 2029.12, Principal Investigator (PI).
National Natural Science Foundation of China (NSFC) Joint Fund Project: "Federated Learning Algorithms for Multimodal Large Model Training", 2026.01 — 2029.12, Participant.

Selected Publications View All Publications →

Fang Peng, Xiaoshan Yang, Linhui Xiao, Yaowei Wang, Changsheng Xu*. SgVA-CLIP: Semantic-Guided Visual Adapting of Vision-Language Models for Few-Shot Image Classification. IEEE Transactions on Multimedia (IEEE TMM), 26: 3469-3480, 2024. CCF-A · CAS Tier-1 TOP IF = 9.7 150+ Citations
Fang Peng, Xiaoshan Yang, Yaowei Wang, Changsheng Xu*. Progressive Learning of Instance-Level Proxy Semantics for Few-Shot Action Recognition. IEEE Transactions on Multimedia (IEEE TMM), 28: 853-864, 2026. CCF-A · CAS Tier-1 TOP IF = 9.7
Fang Peng, Xiaoshan Yang*, Yaowei Wang, Changsheng Xu. Group-Relative Visual Discrimination Enhancement for Unlocking Intrinsic Capability of MLLMs. IEEE Transactions on Circuits and Systems for Video Technology (IEEE TCSVT), 36: 5865-5876, 2026. CAS Tier-1 TOP IF = 11.1
Fang Peng, Zheng Miao, Fei Li, Zhenbo Li*. S-FPN: A shortcut feature pyramid network for sea cucumber detection in underwater images. Expert Systems with Applications (ESWA), 182: 115306, 2021. CAS Tier-1 TOP IF = 7.5 100+ Citations
Linhui Xiao, Xiaoshan Yang, Fang Peng, Yaowei Wang, Changsheng Xu*. OneRef: Unified One-tower Expression Grounding and Segmentation with Mask Referring Modeling. Annual Conference on Neural Information Processing Systems (NeurIPS), 2024. CCF-A
Linhui Xiao, Xiaoshan Yang, Fang Peng, Ming Yan, Yaowei Wang, Changsheng Xu*. CLIP-VG: Self-paced Curriculum Adapting of CLIP for Visual Grounding. IEEE Transactions on Multimedia (IEEE TMM), 26: 4334-4347, 2024. CCF-A · CAS Tier-1 TOP IF = 9.7 100+ Citations

Patents

Authorized Invention Patent: "Few-shot Action Recognition Method, Electronic Device and Medium", Patent No.: ZL 2025 1 0067424.8.
Authorized Invention Patent: "Object Localization Method, Device, Apparatus and Medium", Patent No.: ZL 2024 1 0382411.5.
Authorized Invention Patent: "Object Detection and Segmentation Scheme", Patent No.: ZL 2024 1 0272880.1.

学术成果 (Publications)

论文发表与发明专利（按年份时间轴排列）

2026

Group-Relative Visual Discrimination Enhancement for Unlocking Intrinsic Capability of MLLMs

Fang Peng, Xiaoshan Yang#, Yaowei Wang, Changsheng Xu

IEEE Transactions on Circuits and Systems for Video Technology (IEEE TCSVT), 36: 5865-5876, 2026. 中科院大类一区 TOP JCR-Q1 IF = 11.1

📄 Paper (IEEE) 📥 PDF

@article{peng2026grove,
  title={Group-Relative Visual Discrimination Enhancement for Unlocking Intrinsic Capability of MLLMs},
  author={Peng, Fang and Yang, Xiaoshan and Wang, Yaowei and Xu, Changsheng},
  journal={IEEE Transactions on Circuits and Systems for Video Technology},
  volume={36},
  pages={5865--5876},
  year={2026}
}

Progressive Learning of Instance-Level Proxy Semantics for Few-Shot Action Recognition

Fang Peng, Xiaoshan Yang, Yaowei Wang, Changsheng Xu#

IEEE Transactions on Multimedia (IEEE TMM), 28: 853-864, 2026. CCF-A · 中科院大类一区 TOP JCR-Q1 IF = 9.7

📄 Paper (IEEE) 📥 PDF

@article{peng2026prolips,
  title={Progressive Learning of Instance-Level Proxy Semantics for Few-Shot Action Recognition},
  author={Peng, Fang and Yang, Xiaoshan and Wang, Yaowei and Xu, Changsheng},
  journal={IEEE Transactions on Multimedia},
  volume={28},
  pages={853--864},
  year={2026}
}

2024

SgVA-CLIP: Semantic-Guided Visual Adapting of Vision-Language Models for Few-Shot Image Classification

Fang Peng, Xiaoshan Yang, Linhui Xiao, Yaowei Wang, Changsheng Xu#

IEEE Transactions on Multimedia (IEEE TMM), 26: 3469-3480, 2024. CCF-A · 中科院大类一区 TOP IF = 9.7 Google Scholar 引用 150+

📄 Paper (IEEE) 📥 PDF 🔗 arXiv

@article{peng2024sgvaclip,
  title={SgVA-CLIP: Semantic-Guided Visual Adapting of Vision-Language Models for Few-Shot Image Classification},
  author={Peng, Fang and Yang, Xiaoshan and Xiao, Linhui and Wang, Yaowei and Xu, Changsheng},
  journal={IEEE Transactions on Multimedia},
  volume={26},
  pages={3469--3480},
  year={2024}
}

OneRef: Unified One-tower Expression Grounding and Segmentation with Mask Referring Modeling

Linhui Xiao, Xiaoshan Yang, Fang Peng, Yaowei Wang, Changsheng Xu#

The Thirty-eighth Annual Conference on Neural Information Processing Systems (NeurIPS), 2024. CCF-A 类顶级会议

📄 Paper (NeurIPS) 🔗 arXiv

@inproceedings{xiao2024oneref,
  title={OneRef: Unified One-tower Expression Grounding and Segmentation with Mask Referring Modeling},
  author={Xiao, Linhui and Yang, Xiaoshan and Peng, Fang and Wang, Yaowei and Xu, Changsheng},
  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
  year={2024}
}

HiVG: Hierarchical Multimodal Fine-grained Modulation for Visual Grounding

Linhui Xiao, Xiaoshan Yang, Fang Peng, Yaowei Wang, Changsheng Xu#

Proceedings of the 32nd ACM International Conference on Multimedia (ACM MM), pp. 5460-5469, 2024. CCF-A 类顶级会议

📄 Paper 🔗 arXiv

@inproceedings{xiao2024hivg,
  title={HiVG: Hierarchical Multimodal Fine-grained Modulation for Visual Grounding},
  author={Xiao, Linhui and Yang, Xiaoshan and Peng, Fang and Wang, Yaowei and Xu, Changsheng},
  booktitle={Proceedings of the 32nd ACM International Conference on Multimedia},
  pages={5460--5469},
  year={2024}
}

CLIP-VG: Self-paced Curriculum Adapting of CLIP for Visual Grounding

Linhui Xiao, Xiaoshan Yang, Fang Peng, Ming Yan, Yaowei Wang, Changsheng Xu#

IEEE Transactions on Multimedia (IEEE TMM), 26: 4334-4347, 2024. CCF-A · 中科院大类一区 TOP IF = 9.7 Google Scholar 引用 100+

📄 Paper (IEEE) 🔗 arXiv

@article{xiao2024clipvg,
  title={CLIP-VG: Self-paced Curriculum Adapting of CLIP for Visual Grounding},
  author={Xiao, Linhui and Yang, Xiaoshan and Peng, Fang and Yan, Ming and Wang, Yaowei and Xu, Changsheng},
  journal={IEEE Transactions on Multimedia},
  volume={26},
  pages={4334--4347},
  year={2024}
}

2021

S-FPN: A shortcut feature pyramid network for sea cucumber detection in underwater images

Fang Peng, Zheng Miao, Fei Li, Zhenbo Li#

Expert Systems with Applications (ESWA), 182: 115306, 2021. 中科院大类一区 TOP IF = 7.5 Google Scholar 引用 100+

📄 Paper (Elsevier DOI)

@article{peng2021sfpn,
  title={S-FPN: A shortcut feature pyramid network for sea cucumber detection in underwater images},
  author={Peng, Fang and Miao, Zheng and Li, Fei and Li, Zhenbo},
  journal={Expert Systems with Applications},
  volume={182},
  pages={115306},
  year={2021}
}

PanGu-α: Large-scale Autoregressive Pretrained Chinese Language Models with Auto-parallel Computation

Wei Zeng, Xiaozhe Ren, Teng Su, ..., Fang Peng, ..., Yaowei Wang, Xuefeng Jin, Qun Liu, Yonghong Tian

arXiv preprint arXiv:2104.12369, 2021. Google Scholar 引用 330+ 国产千亿级中文预训练大模型代表作

🔗 arXiv 📥 PDF

@article{zeng2021pangu,
  title={PanGu-alpha: Large-scale Autoregressive Pretrained Chinese Language Models with Auto-parallel Computation},
  author={Zeng, Wei and Ren, Xiaozhe and Su, Teng and others and Peng, Fang and others and Tian, Yonghong},
  journal={arXiv preprint arXiv:2104.12369},
  year={2021}
}

2018

Water quality prediction model combining sparse auto-encoder and LSTM network

Zhenbo Li#, Fang Peng, Bingshan Niu, Guangyao Li, Jing Wu, Zheng Miao

IFAC-PapersOnLine, 51(17): 831-836, 2018. Google Scholar 引用 100+