About Me

I am Jiahao Tang. My research interests include Vision-Centric AI, Agentic Coding, and Vision-Language-Action (VLA) models.

I received my M.Sc. and B.Eng. degrees from Sun Yat-sen University. I am currently a Research Assistant at the CSU-JPG Lab, Central South University, under the supervision of Prof. Alex Jinpeng Wang.

My research primarily focuses on multimodal models, especially Vision-Language-Code Alignment. I am also interested in world models and VLA, and I am actively exploring the intersection of these areas.

🌟🌟 I am actively looking for PhD positions starting in Spring/Fall 2027.

News

2026.06: Our paper FlowInOne: Unifying Multimodal Generation as Image-In Image-Out Flow Matching was accepted to ECCV 2026.
2026.04: Our paper From Charts to Code: A Hierarchical Benchmark for Multimodal Models was accepted to ACL Main 2026.
2026.02: Our paper Residual Decoder Adapter was accepted to CVPR 2026.
2026.07: I joined CSU-JPG Lab as a Research Assistant.
2025.06: I received my M.Eng. degree in Electronic Science and Technology from Sun Yat-sen University.

Experience

Research Assistant, JPG Lab, Central South University

2025.07 – Present
Supervisor: Alex Jinpeng Wang

Conduct research on the understanding, generation, and evaluation of large multimodal models (LMMs), with a focus on Vision-to-Code Generation and Vision-Centric multimodal generation.
From Charts to Code: A Hierarchical Benchmark for Multimodal Models (ACL Main 2026, first author, accepted)
Proposed a user-interactive Chart-to-Code benchmark covering chart reproduction, chart editing, and long-table-to-chart generation; evaluated 29 models and designed a rubric-based evaluation pipeline measuring code executability, code correctness, and visual consistency.
FlowInOne: Unifying Multimodal Generation as Image-in, Image-out Flow Matching (ECCV 2026, first author, under review)
Proposed an image-in, image-out multimodal generation paradigm that converts text descriptions, spatial layouts, and editing instructions into visual prompts; built the VisPrompt-5M dataset and conducted experiments with Qwen-Image-Edit and the CrossFlow framework.
Residual Decoder Adapter: ID-Preserving Tokenizer Adaption for AR Text Rendering (CVPR 2026, co-author, accepted)
Investigated poor text rendering quality in VQ-VAE-based autoregressive unified models, identified visual tokenizer reconstruction as a key bottleneck, and validated RDA through systematic evaluation and ablation studies on Janus-Pro, Lumina-mGPT, and TAR.

Algorithm Intern, Guangdong Engineering Center for IoT Chips and Systems

Supervisor: Jianguo Hu 2022.05 – 2023.09

Worked on software-hardware co-design algorithms for national and provincial research projects, including AI-assisted EDA for automatic analog IC layout optimization.
Built data pipelines for analog circuit netlist parsing, graph-structured modeling, and topology feature extraction to support GNN-based circuit performance prediction and intelligent EDA tasks.
Developed VQA systems based on CLIP, BERT, and BLIP for cross-modal semantic alignment and reasoning.

Algorithm Intern, Guangdong Provincial Key Laboratory of Advanced IC Design and Integration Technology

Supervisor: Jianguo Hu 2023.10 – 2025.03

Built multimodal understanding systems with Qwen-VL, Whisper, and knowledge graph modules for cross-media intelligent understanding.
Supported pedestrian action recognition and VideoQA by processing video data, fine-tuning spatio-temporal representation models, and evaluating model performance.

Selected Publications

* indicates equal contribution.

ECCV 2026

FlowInOne: Unifying Multimodal Generation as Image-in, Image-out Flow Matching
ECCV 2026
Junchao Yi^*, Rui Zhao^*, Jiahao Tang^*, Weixian Lei, Linjie Li, Qisheng Su, Zhengyuan Yang, Lijuan Wang, Xiaofeng Zhu, Alex Jinpeng Wang.

[Project Page][arXiv][Github]

ACL 2026

From Charts to Code: A Hierarchical Benchmark for Multimodal Models
ACL Main 2026
Jiahao Tang^*, Henry Hengyuan Zhao^*, Lijian Wu, Yifei Tao, Dongxing Mao, Yang Wan, Jingru Tan, Min Zeng, Min Li, Alex Jinpeng Wang.

[Project Page][arXiv][Github]

CVPR 2026

Residual Decoder Adapter: ID-Preserving Tokenizer Adaption for Autoregressive Text Rendering
CVPR 2026
Dongxing Mao^*, Alex Jinpeng Wang^*, Jiahao Tang, Kevin Qinghong Lin, Linjie Li, Zhengyuan Yang, Lijuan Wang, Min Li, Jingru Tan.

[Project Page][arXiv][Github]

Full Publications

From Charts to Code: A Hierarchical Benchmark for Multimodal Models, Jiahao Tang^*, Henry Hengyuan Zhao^*, Lijian Wu, Yifei Tao, Dongxing Mao, Yang Wan, Jingru Tan, Min Zeng, Min Li, Alex Jinpeng Wang, ACL Main 2026. [Project Page][arXiv][Github]
Residual Decoder Adapter: ID-Preserving Tokenizer Adaption for Autoregressive Text Rendering, Dongxing Mao^*, Alex Jinpeng Wang^*, Jiahao Tang, Kevin Qinghong Lin, Linjie Li, Zhengyuan Yang, Lijuan Wang, Min Li, Jingru Tan, CVPR 2026. [Project Page][arXiv][Github]
FlowInOne: Unifying Multimodal Generation as Image-in, Image-out Flow Matching, Junchao Yi^*, Rui Zhao^*, Jiahao Tang^*, Weixian Lei, Linjie Li, Qisheng Su, Zhengyuan Yang, Lijuan Wang, Xiaofeng Zhu, Alex Jinpeng Wang, ECCV 2026. [Project Page][arXiv][Github]
ChartJudge: Evaluating Reward Models for LMM-based Chart-to-Code Generation, Lijian Wu^*, Henry Hengyuan Zhao^*, Zijian Zhang^*, Jiahao Tang^*, Jiajun Wu, Alex Jinpeng Wang, under review at ECCV 2026.
Internalize External Competence for Visual Instruction Editing, Wenjun Huang^*, Rui Zhao^*, Suyang Hou^*, Jiahao Tang, Wenjia Wang, Zhuobai Dong, Alex Jinpeng Wang, Hu Jian Guo, under review at NeurIPS 2026.
Enhancing Visual Understanding in Multimodal Large Language Models with Efficient Feature Alignment and State Space Models, Wenjun Huang^*, Jiakai Pan^*, Jiahao Tang, Yifei Xing, Yuhe Wang, Zhengzhuo Wang, Shengzhi Shen, Jianguo Hu, ECAI 2025.
VisCompConText: Scaling Multi-Modal Contexts via Visual Token Compression and Language Model Guidance, Wenjun Huang, Zhengzhuo Wang, Jiakai Pan, Yuhe Wang, Shengzhi Shen, Jiahao Tang, Chaoxing Zhou, Jianguo Hu, ECAI 2025.
Spatio-Temporal Graph Convolution Transformer for Video Question Answering, Jiahao Tang, Jianguo Hu, Wenjun Huang, Shengzhi Shen, Jiakai Pan, Deming Wang, Yanyu Ding, IEEE Access 2024.
Chinese Named Entity Recognition for IC Patent Domain Based on RoBERTa-wwm-ext, GCN and Efficient Global Pointer, Yunxiao Lin^*, Jiahao Tang^*, Wenjun Huang, Yanyu Ding, 2024 5th International Conference on Computing, Networks and Internet of Things.

Education

Sun Yat-sen University

M.Sc. in Electronic Science and Technology, 2022.09 – 2025.06
GPA: 3.8/4.0

Sun Yat-sen University

B.Eng. in Electronic Information Science and Technology, 2018.09 – 2022.06
GPA: 3.5/4.0

Skills

Programming: Python, PyTorch, Verilog, C++, MATLAB, LaTeX.
Hardware: Analog IC Design with Virtuoso, Digital IC Design with ModelSim, Antenna Design with HFSS.
Languages: Mandarin Chinese, English.