Jiahao Lyu

I'm a PhD student at IIE, CAS in Beijing, China. At IIE, I've worked on InTimeLab, IIE, CAS, advised by Professor Yu Zhou and Can Ma. Before that, I got a Bachelor`s degree from Beijing Normal University.

Email / CV / Scholar / Github

News

2025/07: Selected as a Reviewer of AAAI 2026.
2025/06: A co-first-author paper has been accepted by ICDAR 2025.
2025/05: A first-author paper has been accepted by TOMM 2025.
2025/04: A co-first-author paper has been accepted by IJCAI 2025.
2025/03: Selected as a Reviewer of IJCAI 2025.
2025/02: Selected as a Reviewer of ICDAR 2025.
2024/12: A first-author paper has been accepted by AAAI2025.
2024/12: A cooperative paper has been accepted by ICASSP2025.
2023/06: A cooperative paper has been accepted by PRCV2023.
2022/12: Third Prize, Channel of street text detection and recognition, 1st International Algorithm Cases Competition, Pazhou Lab, Huangpu.
2022/06: A cooperative paper has been accepted by ACM MM2022.

Research

I'm interested in Scene Text Detection, Recognition & Spotting. Below are some selected publications.

	The Devil is in Fine-tuning and Long-tailed Problems: A New Benchmark for Scene Text Detection Tianjiao Cao, Jiahao Lyu, Weichao Zeng, Weiming Mu, Yu Zhou, IJCAI, 2025 Code / arXiv We uncover two key factors contributing to this discrepancy through extensive scene text detection experiments, Fine-tuning Gap and long-tailed distribution of texts we advocate for a Joint-Dataset Learning (JDL) protocol to alleviate the Fine-tuning Gap. Additionally, an error analysis is conducted to identify three major categories and 13 subcategories of challenges in long-tailed scene text, upon which we propose a Long-Tailed Benchmark (LTB).
	Arbitrary Reading Order Scene Text Spotter with Local Semantics Guidance Jiahao Lyu, Wei Wang, Dongbao Yang, Jinwen Zhong, Yu Zhou, AAAI, 2025 Code / arXiv We propose LSGSpotter, a local semantics-guided scene text spotter to handle the arbitrary reading order text instances without sophisticated detection.
	Class-Agnostic Region-of-Interest Matching in Document Images Demin Zhang, Jiahao Lyu, Zhijie Shen, Yu Zhou, ICDAR, 2025 Code / arXiv We define a new task, RoI-Matching, to match the customized regions in a flexible, efficient, multi-granular, and open-set manner.
	Char-SAM: Turning Segment Anything Model into Scene Text Segmentation Annotator with Character-level Visual Prompts Enze Xie Jiaho Lyu, Daiqing Wu, Huawen Shen, Yu Zhou, ICASSP, 2025 arXiv We propose an automatic annotation pipeline named Char-SAM, that turns SAM into a low-cost segmentation annotator with a Character-level visual prompt.
	TextBlockV2: Towards Precise-Detection-Free Scene Text Spotting with Pre-trained Language Model Jiaho Lyu, Jin Wei, Gangyan Zeng, Zeng Li, Enze Xie, Wei Wang, Can Ma, Yu Zhou, TOMM, 2025 arXiv Extension of TextBlock (ACM MM2022)