Elamur Research⌕
☰
home / papers / 2026 / Flow-Based Page Unique Semantic Mapping Architecture for Document Visual Question Answering

Flow-Based Page Unique Semantic Mapping Architecture for Document Visual Question Answering

Paper · 2026 · Haosen Wang, Jing Xiao, Chaochao Du et al.

Summary

The paper targets multi-page Document Visual Question Answering (DocVQA). OCR-free generative models still often pick the wrong evidence page when several pages look semantically alike, because they rank pages by coarse question–page similarity. The authors argue formally, through three propositions, that for a given question every page has its own unique semantic representation, and that pages and these semantics are related by a bijection. Building on this, they propose FUMA. Its pipeline has three stages:

  • pretrained image and text encoders embed the pages and the question;
  • a stack of invertible flow-based blocks, combining fine-grained cross-modal attention with likelihood regularization, maps each page to a distinct, question-conditioned semantic vector, which prevents different pages from collapsing onto similar points;
  • a mixture-of-experts decoder produces the answer from the chosen page.

Evidence localization thus becomes a selection over well-separated representations rather than a similarity search. Across four benchmarks, FUMA beats prior methods in both page localization and answer quality. The gains persist across different backbones and against larger MLLMs, which suggests they come from the mapping mechanism rather than model scale.

Key points

  • On DUDE, the 4B FUMA model scores EM 73.72, F1 87.86, ANLS 91.13 and AC (answer-conditioned page localization accuracy) 91.24.
  • It outperforms Qwen2.5-VL (7B), InternVL2.5 (8B) and LLaVA-1.6 (13B). Its AC is 30.67 points higher than the best of these (LLaVA-1.6, 60.57).
  • FUMA was compared with MP-FIRE under three encoder pairs: ViT-Base+BERT-Base, Swin-Base+RoBERTa-Large and Florence-2+Qwen2.5. FUMA won on every metric, with EM gains of +6.33, +3.56 and +3.78.
  • The flow module uses invertible transformations in the style of NICE, RealNVP and Glow, trained with Jacobian log-determinant likelihood terms.
  • Stated limitations:
    • extra compute from the flow computations;
    • a need for page-level evidence supervision during training;
    • no grounding below the page level, such as table cells or chart elements.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Tianjin UniversityTianjin · China41 works · 1,128 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 2Research subjects and languages the work is about.Vision & language · 657 worksQuestion answering · 441 works

Cite

Cite as Haosen Wang, Jing Xiao, Chaochao Du, Xiaowang Zhang and Zhiyong Feng. 2026. Flow-Based Page Unique Semantic Mapping Architecture for Document Visual Question Answering. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.1679

@inproceedings{wang2026flow,
  title = {{Flow-Based Page Unique Semantic Mapping Architecture for Document Visual Question Answering}},
  author = {Haosen Wang, Jing Xiao, Chaochao Du, Xiaowang Zhang and Zhiyong Feng},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1679},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1679}
}

References

46 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. 2017. Density estimation using real nvp. Preprint, arXiv:1605.08803. — [arXiv:1605.08803](https://arxiv.org/abs/1605.08803) find
  2. Srikar Appalaraju, Bhavan Jasani, Bhargava Urala Kota, Yusheng Xie, and R. Manmatha. 2021. Docformer: End-to-end transformer for document understanding. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 993– 1003. find
  3. Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, and 8 others. 2025. Qwen2.5-vl technical report. Preprint, arXiv:2502.13…
  4. Tsachi Blau, Sharon Fogel, Roi Ronen, Alona Golts, Shahar Tsiper, Elad Ben Avraham, Aviad Aberdam, Roy Ganz, and Ron Litman. 2024. GRAM: Global Reasoning for Multi-Page VQA . In 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 15598–15607, Los Alamitos, CA, USA. IEEE… find
  5. Zhe Chen, Weiyun Wang, Yue Cao, Yangzhou Liu, Zhangwei Gao, Erfei Cui, Jinguo Zhu, Shenglong Ye, Hao Tian, Zhaoyang Liu, Lixin Gu, Xuehui Wang, Qingyun Li, Yiming Ren, Zixuan Chen, Jiapeng Luo, Jiahao Wang, Tan Jiang, Bo Wang, and 23 others. 2025. Expanding performance boundaries of open-source mult…
  6. Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. 2021. An image is worth 16x16 words: Transformers for image recognition at scale. In Intern…
  7. Will Grathwohl, Ricky T. Q. Chen, Jesse Bettencourt, Ilya Sutskever, and David Duvenaud. 2018. Ffjord: Free-form continuous dynamics for scalable reversible generative models. Preprint, arXiv:1810.01367. — [arXiv:1810.01367](https://arxiv.org/abs/1810.01367) find
  8. Julian Jorge Andrade Guerreiro, Naoto Inoue, Kento Masui, Mayu Otani, and Hideki Nakayama. 2024. Layoutflow: Flow matching for layout generation. In Computer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part XXXVI, page 56–72, Berlin, Heidelb… find
  9. Jonathan Herzig, Pawel Krzysztof Nowak, Thomas Müller, Francesco Piccinno, and Julian Eisenschlos. 2020. TaPas: Weakly supervised table parsing via pre-training. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4320–4333, Online. Association for Compu…
  10. Anwen Hu, Yaya Shi, Haiyang Xu, Jiabo Ye, Qinghao Ye, Ming Yan, Chenliang Li, Qi Qian, Ji Zhang, and Fei Huang. 2024. mplug-paperowl: Scientific diagram analysis with the multimodal large language model. In Proceedings of the 32nd ACM International Conference on Multimedia, MM ’24, page 6929–6938, N… find
  11. Ronghang Hu, Amanpreet Singh, Trevor Darrell, and Marcus Rohrbach. 2020. Iterative answer prediction with pointer-augmented multimodal transformers for textvqa. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9989–9999. find
  12. Xin Huang and Kye Min Tan. 2025. Beyond text: Unlocking true multimodal, end-to-end rag with tomoro colqwen3. find
  13. Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, and Furu Wei. 2022. Layoutlmv3: Pre-training for document ai with unified text and image masking. In Proceedings of the 30th ACM International Conference on Multimedia, MM ’22, page 4083–4091, New York, NY, USA. Association for Computing Machinery.
  14. Drew A. Hudson and Christopher D. Manning. 2019. Gqa: A new dataset for real-world visual reasoning and compositional question answering. In 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 6693–6702.
  15. Kushal Kafle, Brian Price, Scott Cohen, and Christopher Kanan. 2018. Dvqa: Understanding data visualizations via question answering. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5648–5656. find
  16. Geewook Kim, Teakgyu Hong, Moonbin Yim, JeongYeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, and Seunghyun Park. 2022. Ocr-free document understanding transformer. In Computer Vision – ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Procee…
  17. Durk P Kingma and Prafulla Dhariwal. 2018. Glow: Generative flow with invertible 1x1 convolutions. In Advances in Neural Information Processing Systems, volume 31. Curran Associates, Inc. find
  18. Jordy Van Landeghem, Rafał Powalski, Rubèn Tito, Dawid Jurkiewicz, Matthew Blaschko, Łukasz Borchmann, Mickaël Coustaty, Sien Moens, Michał Pietruszka, Bertrand Ackaert, Tomasz Stanisławek, Paweł Józiak, and Ernest Valveny. 2023. Document understanding dataset and evaluation (dude). In 2023 IEEE/CVF… find
  19. Chen-Yu Lee, Chun-Liang Li, Timothy Dozat, Vincent Perot, Guolong Su, Nan Hua, Joshua Ainslie, Renshen Wang, Yasuhisa Fujii, and Tomas Pfister. 2022. Formnet: Structural encoding beyond sequential modeling in form document information extraction. Preprint, arXiv:2203.08411. find
  20. Kenton Lee, Mandar Joshi, Iulia Turc, Hexiang Hu, Fangyu Liu, Julian Eisenschlos, Urvashi Khandelwal, Peter Shaw, Ming-Wei Chang, and Kristina Toutanova. 2023. Pix2struct: screenshot parsing as pretraining for visual language understanding. In Proceedings of the 40th International Conference on Mach… find
  21. Kun Li, George Vosselman, and Michael Ying Yang. 2025. Multimodal rationales for explainable visual question answering. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 191–201. find
  22. Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. 2023. Flow matching for generative modeling. In The Eleventh International Conference on Learning Representations.
  23. Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. 2024. Llava-next: Improved reasoning, ocr, and world knowledge. find
  24. Xingchao Liu, Chengyue Gong, and Qiang Liu. 2022. Flow straight and fast: Learning to generate and transfer data with rectified flow. ArXiv, abs/2209.03003.
  25. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Roberta: A robustly optimized bert pretraining approach.
  26. Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. 2021. Swin transformer: Hierarchical vision transformer using shifted windows. Preprint, arXiv:2103.14030.
  27. Yubo Ma, Yuhang Zang, Liangyu Chen, Meiqi Chen, Yizhu Jiao, Xinze Li, Xinyuan Lu, Ziyu Liu, Yan Ma, Xiaoyi Dong, Pan Zhang, Liangming Pan, Yu-Gang Jiang, Jiaqi Wang, Yixin Cao, and Aixin Sun. 2024. MMLONGBENCH-DOC: Benchmarking long-context document understanding with visualizations. In The Thirty-e… find
  28. Minesh Mathew, Dimosthenis Karatzas, and C. V. Jawahar. 2021. Docvqa: A dataset for vqa on document images. In 2021 IEEE Winter Conference on Applications of Computer Vision (WACV), pages 2199–2208.
  29. Armineh Nourbakhsh, Siddharth Parekh, Pranav Shetty, Zhao Jin, Sameena Shah, and Carolyn Rose. 2025. Where is this coming from? making groundedness count in the evaluation of document VQA models. In Findings of the Association for Computational Linguistics: NAACL 2025, pages 5341–5361, Albuquerque,… find
  30. Jinhui Pang, Xinyun Yang, Xiaoyao Qiu, Zixuan Wang, and Taisheng Huang. 2024. Mmaf: Masked multimodal attention fusion to reduce bias of visual features for named entity recognition. DATA INTELLI-GENCE, 6(4):1114–1133. find
  31. Yang Qin, Huiming Xie, Yujie Li, Benying Tan, and Shuxue Ding. 2025. Enhancing intermodal interaction for unified vision-language understanding and generation. DATA INTELLIGENCE, 7(2):358–380. find
  32. Qwen, :, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, and 25 others. 2025. Qwen2.5 technical report. Preprint, arXiv:2412.15115. find
  33. Danilo Jimenez Rezende and Shakir Mohamed. 2015. Variational inference with normalizing flows. In Proceedings of the 32nd International Conference on Machine Learning - Volume 37, ICML’15, page 1530–1538. JMLR.org. find
  34. Johannes Schusterbauer, Ming Gui, Frank Fundel, and Björn Ommer. 2025. Diff2flow: Training flow matching models via diffusion model alignment. In 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 28347–28357. find
  35. Shubhankar Singh, Purvi Chaurasia, Yerram Varun, Pranshu Pandya, Vatsal Gupta, Vivek Gupta, and Dan Roth. 2024. FlowVQA: Mapping multimodal logic in visual question answering with flowcharts. In Findings of the Association for Computational Linguistics: ACL 2024, pages 1330–1350, Bangkok, Thailand.… find
  36. Mohamed Ali Souibgui, Changkyu Choi, Andrey Barsky, Kangsoo Jung, Ernest Valveny, and Dimosthenis Karatzas. 2025. DocVXQA: Context-aware visual explanations for document question answering. In Forty-second International Conference on Machine Learning. find
  37. Bin Xiao, Haiping Wu, Weijian Xu, Xiyang Dai, Houdong Hu, Yumao Lu, Michael Zeng, Ce Liu, and Lu Yuan. 2024. Florence-2: Advancing a unified representation for a variety of vision tasks. In 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4818–4829. find
  38. Tianyu Xie, Yu Zhu, Longlin Yu, Tong Yang, Ziheng Cheng, Shiyue Zhang, Xiangyu Zhang, and Cheng Zhang. 2024. Reflected flow matching. In Proceedings of the 41st International Conference on Machine Learning, ICML’24. JMLR.org. find
  39. Chen Xu, Xiuyuan Cheng, and Yao Xie. 2026. Local flow matching generative models. IEEE Transactions on Information Theory, 72(5):3339–3358. find
  40. Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, and Ming Zhou. 2020. Layoutlm: Pre-training of text and layout for document image understanding. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD ’20, page 1192–1200, New York, NY, USA. A…
  41. Jiabo Ye, Anwen Hu, Haiyang Xu, Qinghao Ye, Ming Yan, Guohai Xu, Chenliang Li, Junfeng Tian, Qi Qian, Ji Zhang, Qin Jin, Liang He, Xin Lin, and Fei Huang. 2023. UReader: Universal OCR-free visually-situated language understanding with multimodal large language model. In Findings of the Association f…
  42. Yongqi Yu, Jinxu Zhang, and Yu Zhang. 2025. Mp-fire: An end-to-end cross-modal framework for complex multi-page document question answering. In 2025 IEEE International Conference on Multimedia and Expo (ICME), pages 1–6. find
  43. Parth Thakkar, Ankush Agarwal, Prasad Kasu, Pulkit Bansal, and Chaitanya Devaguptapu. 2025. Finding needles in images: Can multi-modal llms locate fine details? In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 23626–23648. find
  44. Rubèn Tito, Dimosthenis Karatzas, and Ernest Valveny. 2023. Hierarchical multimodal transformers for mul-tipage docvqa. Pattern Recognition, 144:109834. find
  45. Alexander Tong, Nikolay Malkin, Guillaume Huguet, Yanlei Zhang, Jarrid Rector-Brooks, Fatras Kilian, and Guy Wolf. 2023. Conditional flow matching: Simulation-free dynamic optimal transport. find
  46. Yiwen Wang, Xiaobing Zhao, Xiaoke Qi, Bo Chen, Chuanlian Ma, and Yang Xu. 2025. A large language model evaluation method for legal case retrieval. DATA INTELLIGENCE, 7(2):440–460. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-flow-based-page-unique-semantic/