About me
I am Haiwen Huang, my Chinese name is 黄海文. I am a PhD student at University of Tübingen, co-supervised by Prof. Andreas Geiger and Dr. Dan Zhang, starting in Jan. 2022. I am also an ELLIS PhD student.
My PhD research focuses on improving the generalization of vision and multimodal models, enabling them to perform robustly across diverse tasks and domains. For example, I have developed methods that leverage 3D priors to enhance 2D object detection in GOOD (ICLR 2023) and use self-distillation to upsample features in Vision Foundation Models in LoftUp (coming soon). I have also contributed to building more reliable evaluations of open-vocabulary generalization in RENOVATE (NeurIPS 2024).
I envision my work through a three-tiered “research pyramid”: (1) Method Development, (2) Performance Evaluation, and (3) Normative Meta-Evaluation. My PhD projects focus on the first two levels, but my long-term goal is to address all three to ensure that AI truly benefits society.
🔴 I am curently seeking internships starting from Sept/Oct 2025 onward where I can explore methods for understanding and evaluating the generalization and capabilities of VLMs, as well as the scientific principles behind designing robust benchmarks for them.
Before my PhD studies, I completed my MSc in CS at University of Oxford, advised by Prof. Yarin Gal, and my undergrad study in Mathematics at Peking University, advised by Prof. Bin Dong. I have also worked as a researcher for a year at Megvii (previously known as Face++) with Xinyu Zhou as my group leader, studying large-scale annotation and OoD detection methods.
My CV is here.
Highlighted Research
- Renovating Names in Open-Vocabulary Segmentation Benchmarks
- Haiwen Huang, Songyou Peng, Dan Zhang, Andreas Geiger
- NeurIPS 2024.
- OpenReview, Project Page
- Multimodal Dataset Upgrading: a New Challenge for Data Annotation
- Haiwen Huang, Dan Zhang, Andreas Geiger
- ICLR 2024 DPFM workshop
- Paper
- GOOD: Exploring Geometric Cues for Detecting Objects in an Open World
- Haiwen Huang, Andreas Geiger, Dan Zhang
- ICLR 2023.
- OpenReview, Code
- Decomposing Representations for Deterministic Uncertainty Estimation
- Haiwen Huang, Joost van Amersfoort, Yarin Gal
- 6th Bayesian Deep Learning workshop at NeurIPS 2021.
- Poster
- Feature Space Singularity for Out-of-distribution Detection
Teaching
- Computer Vision (summer term 2023)
- Seminar: Large-scale Generative Models: Prospects and Limitations (summer term 2023)