Yuxuan Luo
E-mail: 2401112141 (at) stu (dot) pku (dot) edu (dot) cn
I am Yuxuan Luo(罗宇轩), a second year Ph.D. student at Wangxuan Institute of Computer Science (WICT), Peking University, advised by Prof. Zhouhui Lian. I received my Bachelor’s degree in Artificial Intelligence from Yuanpei College, Peking University.
My research focuses on understanding and generating knowledge-intensive visual media. I study how images carry factual, disciplinary, and cultural information:
- how to interpret dense knowledge images,
- how to make generative models convey that knowledge precisely, and
- how to measure the faithfulness and clarity of generated visuals.
This work connects unified multimodal understanding and generation with model reasoning and the broader question of whether generative systems function as world-models.
Technically, I focus on multimodal LLMs (mLLMs), diffusion and autoregressive image models, and post-training paradigms such as fine-tuning, LoRA/instruction tuning, and evaluation pipelines. I am actively seeking collaborations — if you’re interested in working with me on knowledge images, please contact me by e-mail.
news
| Sep 19, 2025 | MMMG was accepted by Neurips 2025! MMMG is a large-scale discipline-image benchmark designed to assess text-to-image (T2I) models on their ability to generate faithful and readable visuals. Our cases span 10 disciplines and 6 educational levels. |
|---|---|
| Jun 25, 2025 | CalliReader: Contextualizing Chinese Calligraphy via an Embedding-aligned Vision Language Model was accepted by ICCV 2025! This paper proposes CalliReader, a novel VLM that solves Chinese Calligraphy Contextualization. We also release the first page-level Calligraphy dataset and CalliBench. |
| Apr 06, 2024 | CalliRewrite: Recovering Handwriting Behaviors from Calligraphy Images without Supervision has been selected as a finalist for the IEEE ICRA 2024 Best Paper Award in Service Robotics! This research was conducted during my undergraduate studies under the guidance of Prof. Zhouhui Lian. |
