Region-Wise Correspondence Prediction between Manga Line Art Images
Yingxuan Li, Jiafeng Mao, Qianru Qiu, and Yusuke Matsui
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026
李 映萱 / リ インシュアン
Yingxuan LI
Machine Learning Engineer · Ph.D.
I am a Machine Learning Engineer at CyberAgent in Tokyo, Japan. I received my Ph.D. from the University of Tokyo under the supervision of Prof. Yusuke Matsui. My work focuses on computer vision, vision-language models, and multimodal learning, with a particular interest in comic/manga understanding.
Research Interests
Languages
I am currently open to ML Engineer, Research Engineer, and Research Scientist opportunities in computer vision and multimodal AI.
Picked Publications
Noisy Label Refinement with Semantically Reliable Synthetic Images
Yingxuan Li, Jiafeng Mao, and Yusuke Matsui
IEEE International Conference on Image Processing (ICIP), 2025
Zero-Shot Character Identification and Speaker Prediction in Comics via Iterative Multimodal Fusion
Yingxuan Li, Ryota Hinami, Kiyoharu Aizawa, and Yusuke Matsui
ACM International Conference on Multimedia (ACM MM), Oral, 2024
Manga109Dialog: A Large-scale Dialogue Dataset for Comics Speaker Detection
Yingxuan Li, Kiyoharu Aizawa, and Yusuke Matsui
IEEE International Conference on Multimedia and Expo (ICME), Oral, 2024
Research & Industry Experience
CyberAgent, Inc.
Machine Learning Engineer · AI Localization Center
- Develop multimodal AI systems for manga understanding and localization
- Design translation evaluation frameworks combining LLM-based evaluation with human review
- Collaborate with researchers and engineers to bridge research prototypes and production systems
CyberAgent, Inc.
Research Intern · AI Lab
- Conducted research on visual representation learning for structured image understanding
- Designed a deep learning framework for region-wise correspondence prediction between manga line-art images
- Published the resulting work at CVPR 2026
Mantra Inc.
Research Intern
- Conducted research on multimodal reasoning and vision-language understanding
- Developed a training-free method integrating visual, textual, and contextual information
- Published the resulting work at ACM Multimedia 2024, where it was selected as an Oral paper
Education
The University of Tokyo
Ph.D. in Information Science and Technology
Advisor: Prof. Yusuke Matsui
The University of Tokyo
Master of Information Science and Technology
Advisor: Prof. Yusuke Matsui
Tianjin University
Bachelor of Engineering, Communication Engineering
Advisor: Prof. Yan Xu
- Suzhou Yucai Scholarship, 2018
- Tianjin People's Government Scholarship, 2017
- Merit Student of Tianjin University, 2017, 2018
- Excellent Communist Youth League Cadres of Tianjin University, 2017, 2018
Yokohama National University
Undergraduate Exchange Student
Advisor: Prof. Ryuji Kohno
Other Publications
Region-Level Correspondence Prediction between Manga Line Art Images
Yingxuan Li, Qianru Qiu, Jiafeng Mao, and Yusuke Matsui
画像の認識・理解シンポジウム (MIRU), 2025
Noisy Label Refinement via Reference-Guided Synthetic Images
Yingxuan Li, Jiafeng Mao, and Yusuke Matsui
画像の認識・理解シンポジウム (MIRU), 2025
Iterative Multimodal Fusion を用いたゼロショットの漫画キャラクター識別と話者推定
李映萱, 日並遼太, 相澤清晴, 松井勇佑
言語処理若手シンポジウム (YANS), 2024
Zero-Shot Character Identification and Speaker Prediction in Comics via Iterative Multimodal Fusion
Yingxuan Li, Ryota Hinami, Kiyoharu Aizawa, and Yusuke Matsui
画像の認識・理解シンポジウム (MIRU), 2024
Manga109Dialog: 漫画話者推定のための大規模な対話データセット
李映萱, 相澤清晴, 松井勇佑
コミック工学研究会, 2023
Comic Speaker Detection and Estimation using Scene Graph
Yingxuan Li, Kiyoharu Aizawa, and Yusuke Matsui
パターン認識・メディア理解研究会 (PRMU), 2023
Speaker Detection of Comics using Scene Graph
Yingxuan Li and Yusuke Matsui
画像の認識・理解シンポジウム (MIRU), 2022
Invited Talks
大規模言語モデルとマルチモーダルデータの融合による漫画セリフの話者同定
コミック工学シンポジウム 2024
Manga109Dialog: A Large-scale Dialogue Dataset for Comics Speaker Detection
HUAWEI TOKYO RESEARCH CENTER Discussion Dinner in MIRU 2023