Guian Fang
I am a Ph.D. student at Show Lab, National University of Singapore, advised by Prof. Mike Zheng Shou.
Previously, I received my B.Eng. in HCPLab, Artificial Intelligence at the School of Intelligent Systems Engineering, Sun Yat-sen University, advised by Xiaodan Liang (梁小丹), co-supervised by Shengcai Liao.
Additionally, I was a research scientist at Cybever AI in Sunnyvale, where I cooperated with a small team that mostly works on 3D computer vision and Large language model.
Research
I'm interested in computer vision, deep learning, generative AI, and AI alignment. Most of my research will be about video understanding and generation. Representative papers are highlighted.
|
Generate 3D Worlds in Production with AI
Cybever*, Guian Fang* Image-to-3D: Let AI do the heavy lifting so 3D Professionals can do the storytelling |
|
|
HumanRefiner: Benchmarking Abnormal Human Generation and Refining with Coarse-to-fine Pose-Reversible Guidance
Guian Fang*, Wenbiao Yan*, Yuanfan Guo*, Jianhua Han, Zutao Jiang, Hang Xu, Shengcai Liao, Xiaodan Liang ECCV, 2024 Dataset page / Paper In this project, we introduce AbHuman, the first large-scale benchmark focused on anatomical anomalies. The benchmark consists of 56K synthesized human images, each annotated with 147K human anomalies in 18 different categories. Based on this, we developed HumanRefiner, a novel plug-and-play method for coarse-to-fine refinement of human anomalies. |
|
ChartThinker: A Contextual Chain-of-Thought Approach to Optimized Chart Summarization
Mengsha Liu, Daoyuan Chen, Yaliang Li, Guian Fang, Ying Shen LREC-Coling, 2024 Dataset page / Paper In this project, we address the challenges of chart summarization through our development of ChartThinker, an innovative method that leverages natural language processing to enhance visual-language matching and reasoning capabilities. Our approach involves a large-scale dataset featuring diverse chart-caption pairs and detailed fine-tuning instructions, significantly improving training data effectiveness. ChartThinker employs thought chains and context retrieval strategies to produce logically coherent and accurate summaries. Demonstrating superior performance, our model outperforms eight state-of-the-art models across seven evaluation metrics. |
|
LLaMA2-Accessory: An Open-source Toolkit for LLM Development
Chris Liu, Ziyi Lin, Guian Fang, Jiaming Han, Yijiang Liu, Renrui Zhang, Longtian Qiu, Yichi Zhang, Siyuan Huang 
Checkpoint page / Document LLaMA2-Accessory is an open-source toolkit for pretraining, finetuning and deployment of Large Language Models (LLMs) and multimodal LLMs. This repo is mainly inherited from LLaMA-Adapter with more advanced features. |
Honors & Awards
- 2nd Place, Asia and Pacific Mathematical Contest in Modeling (2022)
- Silver Medal, China Collegiate Algorithm Design & Programming Challenge Contest (2022)
- 2nd Place, Social Computing Innovation Competition (2022)
- 1st Place, National College Computer Ability Challenge (2023)
- Recipient, Huawei Intelligent Foundation Scholarship (2022)
- Recipient, National Encouragement scholarship (2022)
- 1st Place, SYSU Outstanding Student Scholarship (2023)
- Recipient, National Scholarship (2023)
- Recipient, Li Xuerou Foundation Scholarship (2023)
Activities & Services
- Reviewer for ECCV
- Reviewer for NeurIPS
- Reviewer for ICLR
- Reviewer for ICML
- Reviewer for AISTATS
- Reviewer for ACL Rolling Review (ARR), serving conferences including ACL, EMNLP, NAACL, and EACL
- Core organizer of LOVEU: LOng-form VidEo Understanding Towards Multimodal AI Assistant and Copilot Workshop @ CVPR'24
Acknowledgements
I feel incredibly fortunate to have collaborated with such remarkable individuals who have generously offered me their mentorship.
|
|
|