Here is Lin Li (李琳).

I am a postdoctoral fellow in AI Chip Center for Emerging Smart Systems (ACCESS) at the Hong Kong University of Technology and Science (HKUST), advised by Prof. Kwang-Ting (Tim) Cheng. Additionally, I collaborate with Prof. Long Chen at HKUST. Prior to this, I obtained my PhD degree in Computer Science and Technology from Zhejiang University (ZJU), under the supervision of Prof. Jun Xiao.
My research interest includes Multi-modal Large Language Models and Scene Understanding.

If you are interested in any aspect of me, I am always open to discussions and collaborations. Feel free to reach out to me at - lllidy[at]ust.hk

📝 Publications

ACM MM 2025
sym

Compositional zero-shot learning via progressive language-based observations

Lin Li, Guikun Chen, Zhen Wang, Jun Xiao, Long Chen

  • Automatically allocating the observation order in the form of primitive concepts or graduated descriptions, enabling effective prediction of unseen state-object compositions.
TPAMI 2024
sym

Nicest: Noisy label correction and training for robust scene graph generation

Lin Li, Jun Xiao, Hanrong Shi, Hanwang Zhang, Yi Yang, Wei Liu, Long Chen

Code

  • An out-of-distribution scene graph generation dataset VG-OOD and a debiased Knowledge distillation strategy.
NeurIPS 2023
sym

Zero-shot visual relation detection via composite visual cues from large language models

Lin Li, Jun Xiao, Guikun Chen, Jian Shao, Yueting Zhuang, Long Chen

Code

  • The first exploration of zero-shot visual relation detection via composite description prompts.
ICCV 2023
sym

Compositional feature augmentation for unbiased scene graph generation

Lin Li, Guikun Chen, Jun Xiao, Yi Yang, Chunping Wang, Long Chen

Code

  • Tackling unbiased scene graph generation from the perspective of increasing the diversity of triplet features.
CVPR 2022 Oral
sym

The devil is in the labels: Noisy label correction for robust scene graph generation

Lin Li, Long Chen, Yifeng Huang, Zhimeng Zhang, Songyang Zhang, Jun Xiao

Code

  • Reformulating scene graph generation (SGG) as a noisy label learning problem, and pointing out that the two plausible assumptions are not applicable for SGG.

(†: Corresponding author, *: Equal contribution, ♢: Student first author)

🎖 Honors and Awards

  • 2023.10 Academic Star Training Program for Doctoral Students in Zhejiang University.
  • 2022.09 Transfar Group Scholarship at Zhejiang University.
  • 2021.09 National Scholarship.
  • 2017.09 National Scholarship.