COMPUTER VISION & MULTIMODAL LEARNING

Yingjin Li.

李映进

Seeing beyond
the expected.

I study open-world anomaly detection, exploring multimodal learning and large vision-language models for generalizable detection and interpretable reasoning.

Yingjin Li 李映进 — portrait
CURRENTLY ATHKUST (Guangzhou)M.Phil. · Data Science and AnalyticsSupervisor: Prof. Lei ZhuLet’s connect

01 / THE PUBLICATIONS

Publication List.

A record of the questions
I have been working on.

3 papers
  1. ROAD: Rule-Grounded Context-Aware Open-World Driver Anomaly Detection

    Yingjin Li, Shuaibo Li, Wei Ma, Zhijie Qiu, Fuxiang Zhai, Sixiang Chen, Zhaolu Kang, and Lei Zhu.

    Advances in Neural Information Processing Systems (NeurIPS), 2026Accepted
  2. Fashion Image Retrieval Based on Multimodal Features Enhancement and Fusion

    Yingjin Li, Shufan He, Zhaojing Wang, Jin Huang, Xinrong Hu, and Li Li.

    Computer Animation and Social Agents (CASA 2024)Published

    Springer, CCIS 2374, pp. 118–131, 2025.

  3. PriSM-AD: Learning Common Normal Priors and Anomaly Semantics for Diffusion-Based Anomaly Detection

    Yingjin Li, Sixiang Chen, Wenxue Li, Shuaibo Li, Fugee Tsung, and Lei Zhu.

    IEEE Transactions on Multimedia (TMM)Under review

02 / INSIDE THE RESEARCH

A closer look.

Two ways to make
the unexpected understandable.

01NeurIPS 2026 · Accepted
Full citation

RULES · CONTEXT · UNDERSTANDING

ROAD.

The same action.
A different meaning.

Integrating driver video, driving context, and behavioral rules to identify previously unseen driver anomalies—and explain why a behavior is abnormal.

Multimodal LLMsOpen-world reasoning
Research contributions
  • Designed rule-grounded reasoning that translates domain knowledge into structured guidance for multimodal large language models.
  • Integrated context-aware visual reasoning to distinguish similar behaviors across driving situations and account for context-dependent safety rules.
  • Built an evaluation pipeline across driver-monitoring datasets with free-form anomaly descriptions, human assessments, and MLLM baselines.
A THOUGHT EXPERIMENTCONTEXT MATTERS
Illustrative driver looking at a phone, with a switchable parked or driving context P
SAFELY PARKED
OBSERVATION Looking at a phone
Change the context
CONTEXT-APPROPRIATE

When safely parked, looking at a phone does not conflict with the active task of driving.

Illustrative scenario · manually defined responses, not model predictions.

02TMM · Under review
Full citation

NORMAL PRIORS · ANOMALY SEMANTICS

PriSM-AD.

Imagine the normal.
Reveal the anomaly.

Combining common normal visual priors with anomaly semantics in a two-stage diffusion framework for defect detection and pixel-level localization.

Diffusion modelsIndustrial vision
Research contributions
  • Developed prior-guided reconstruction to recover normal appearance while suppressing anomalous structures.
  • Combined semantic anomaly representations with reconstruction cues to identify subtle defects across industrial categories.
  • Analyzed reconstruction quality, cross-category generalization, anomaly localization, and inference efficiency.
A VISUAL EXPLANATION
Illustrated metal ring with a smooth, regular surface
The same illustrated ring with an added surface scratch
ILLUSTRATED INPUTNORMAL PRIOR
← Drag to compare →Input ↔ Normal prior
01 Observe02 Reconstruct03 Localize

Conceptual artwork · illustrates the reconstruction idea, not experimental results.

Illustrated fashion image cards representing multimodal feature fusion
CASA 2024 · Published

Connecting visual features
through multimodal learning.

Fashion Image Retrieval Based on Multimodal Features Enhancement and Fusion

Read publication

03 / THE JOURNEY

Academic roots.

SEP 2025 — JUN 2027 EXPECTED

The Hong Kong University of Science and Technology (Guangzhou)

M.Phil. · Data Science and Analytics

Supervisor: Prof. Lei Zhu
Red Bird Scholarship recipient

SEP 2021 — JUN 2025

Wuhan Textile University

B.Eng. · Data Science and Big Data Technology

Average score: 86.7/100 · Top 5% academically
Overall ranking: 1st in major

04 / RECOGNITION

Selected honors.

  • National Scholarship

    Ministry of Education of China · 2024

  • First Prize, Hubei Province

    China Mathematical Contest in Modeling

  • National First Prize

    Teddy Cup Data Mining Challenge

  • National First Prize

    Huazhong Cup Mathematical Modeling Challenge

05 / LET’S CONNECT

Good research begins
with a conversation.

For research discussions and academic opportunities.

yli499@connect.hkust-gz.edu.cn
Download CV Academic CV · PDFEnglish proficiency · IELTS 6.5