👋 About me

I am Yifan Liang, a graduate student pursuing M.S. in Intelligent Science and Technology at School of Artificial Intelligence and Automation (AIA), Huazhong University of Science and Technology (HUST). I graduated with honors from HUST in 2025, receiving the Outstanding Undergraduate Thesis Award and being recognized as an Honored Graduate.

My research interests focus on computer vision and machine learning, particularly in vision-language models(VLMs) and out-of-distribution (OOD) detection. I am passionate about developing robust AI systems that can handle real-world challenges and uncertainties.

I have published research at top-tier AI conferences, including a Highlight paper @ CVPR 2025 (top 2.98%), which addresses the shortcut problem in VLMs for robust OOD detection. I was also awarded the National First Prize in the 18th “Challenge Cup” National College Student Extracurricular Academic and Technological Works Competition.

📖 Educations

• 2021.09 - 2025.06, B.S. in Artificial Intelligence @ AIA, HUST, Wuhan, China.
• 2025.09 - 2028.06 (expected), M.S. in Intelligent Science and Technology @ AIA, HUST, Wuhan, China.

🎖 Honors and Awards

• 2024.11 National First Prize of the 18th “Challenge Cup” National College Student Extracurricular Academic and Technological Works Competition
• 2025.06 Honored Graduate of HUST
• 2025.06 Outstanding Undergraduate Thesis of HUST

🔥 News

• 2025.06:  🎉🎉 I was awarded Honored Graduate and Outstanding Undergraduate Thesis by HUST.
• 2025.03:  🎉🎉 One paper has been selected as a Highlight @ CVPR 2025 (top 2.98%) .

📝 Publications

arxiv 2025

OpenHAIV: A Framework Towards Practical Open-World Learning

Xiang Xiang^12†, Qinhao Zhou¹, Zhuo Xu¹, Jing Ma¹, Jiaxin Dai¹, Yifan Liang¹, Hanlin Li¹

¹School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan, China

²Peng Cheng Laboratory, Shenzhen, China

^†Corresponding author

Paper | Doc | Code

• Submitted to MVA. This work presents the OpenHAIV framework, enabling unified evaluation across multiple tasks such as Out-of-Distribution Detection, Novel Class Discovery, and Incremental Learning. We hope this framework can bridge the entire pipeline of open-world learning.

arxiv 2025

OpenEarthSensing: Large-Scale Fine-Grained Benchmark for Open-World Remote Sensing

Xiang Xiang^12†, Zhuo Xu¹, Yao Deng¹, Qinhao Zhou¹, Yifan Liang¹, Ke Chen², Qingfang Zheng², Yaowei Wang², Xilin Chen³, Wen Gao²

¹School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan, China

²Peng Cheng Laboratory, Shenzhen, China

³Chinese Academy of Sciences, Beijing, China

^†Corresponding author

Paper | Project | Dataset

• Submission to IJCV. This work proposes a large-scale fine-grained open-world remote-sensing datasets and benchmark OpenEarthSensing(OES), which supports evaluation on multiple tasks including Out-of-Distribution Detection, Class-Incremental Learning, Coarse-to-Fine Few-Shot Class-Incremental Learning and Domain-Incremental Learning.

CVPR 2025

Overcoming Shortcut Problem in VLM for Robust Out-of-Distribution Detection

Zhuo Xu^1*, Xiang Xiang^1*†, Yifan Liang¹

¹School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan, China

^*Equal contribution

^†Corresponding author

Paper | Project | Code | Dataset

• Highlight Paper @ CVPR 2025. This work addresses the shortcut problem in VLMs for robust OOD detection and propose a new OOD dataset ImageNet-Bg.

🚀 Projects

OpenHAIV : A Framework Towards Practical Open-World Learning

• Description: This is a framework for open-world object recognition, supporting out-of-distribution detection, novel class discovery, and incremental learning algorithms to enable robust and flexible object recognition in unconstrained environments.
• Highlights:
  • 🧩 Modular Design
  • 🚀 Scalable Architecture
  • 🔌 Extensible Framework
• Links: Code | Doc

2025 Advanced Machine Learning Course Project: CLIP-based Few-shot OOD Detection

• Description: Focused on CLIP-based few-shot OOD detection, implementing various prompt learning methods including CoOp, LoCoOp, and SCT.
• Tech Stack: CLIP, Prompt Learning, OOD Detection
• Highlights:
  • Implemented and improved three different prompt learning methods for OOD detection
  • Achieved significant improvements in AUROC and FPR95 compared to baseline on ImageNet OOD Benchmark
• Links: Code

2024 NLP Course Project: Multi-modal Image Description with MiniCPM-V

• Description: Implementing image description generation using MiniCPM-V model with both LoRA and full fine-tuning approaches.
• Tech Stack: MLLM, LoRA, Gradio
• Highlights:
  • Implemented both LoRA and full fine-tuning strategies on MiniCPM-V
  • Created interactive web demo for real-time image description generation
• Links: Code

💻 Internships

• Open to internship opportunities in artificial intelligence research and engineering, with strong foundation in machine learning, computer vision, and programming skills (Python, PyTorch, etc.). Feel free to contact me 😊