pith. sign in

arxiv: 2606.23539 · v1 · pith:Z6SUWS4Onew · submitted 2026-06-22 · 💻 cs.CV

LightSTAR: Efficient Visual Document Retrieval via Lightweight Selection with Vision-Adaptive Refinement

Tongkun Guan , Haocheng Wang , Wei Shen , Xiaokang Yang This is my paper

Pith reviewed 2026-06-26 08:41 UTC · model grok-4.3

classification 💻 cs.CV
keywords visual document retrievalefficient multi-modal retrievalcandidate selectionsemantic refinementlatency reductioncontrastive learningregion-wise feature fusion
0
0 comments X

The pith

LightSTAR splits visual document retrieval into fast keyword-based candidate selection and targeted refinement to deliver top accuracy at several-fold lower latency.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper seeks to overcome the high cost of running multi-modal large language models on every page when retrieving relevant visual documents from large collections. It rests on the observation that typical queries contain specific words that appear directly in the text of the pages that matter. LightSTAR therefore first uses simple, LLM-free methods to encode the query and page visuals, quickly producing a small high-recall set of candidates. Only those candidates then receive a more expensive but precise refinement step that fuses text and layout features region by region. The result is retrieval accuracy that matches or exceeds current best methods while the overall time cost drops sharply.

Core claim

LightSTAR decomposes visual document retrieval into an LLM-free Visual Selection stage, which applies content-grounded query encoding and LLM-free visual embeddings to produce a high-recall candidate set, followed by a Vision-adaptive Semantic Refinement stage that performs fine-grained semantic matching exclusively on those candidates through adaptive region-wise feature fusion and a hardness-aware contrastive objective. This yields state-of-the-art retrieval accuracy while reducing end-to-end latency by several-fold.

What carries the argument

Two-stage pipeline of LLM-free Visual Selection for rapid high-recall filtering followed by Vision-adaptive Semantic Refinement using adaptive region-wise feature fusion on the selected candidates only.

If this is right

  • The selection stage can filter thousands of pages using only lightweight embeddings without invoking heavy models on every page.
  • Restricting the refinement stage to a small candidate set directly lowers total computation while preserving accuracy.
  • Adaptive region-wise fusion of textual and layout cues improves matching quality beyond uniform page-level features.
  • The hardness-aware contrastive objective focuses training on difficult negatives to raise final ranking precision.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same lightweight-first decomposition could apply to other large-scale multi-modal search settings where initial keyword signals exist.
  • If the selection stage maintains high recall at scale, the approach would make retrieval feasible over corpora orders of magnitude larger than current MLLM-only pipelines allow.
  • Designers of document systems might deliberately encourage queries that contain distinctive visible words to maximize the efficiency gain.
  • Replacing the visual encoder in the selection stage with even lighter alternatives could be tested without retraining the refinement stage.

Load-bearing premise

User queries are typically keyword-anchored, with semantically rich words expected to appear directly in the visible text of relevant pages.

What would settle it

Measure recall of the LLM-free selection stage on a test set where all queries have been rewritten to avoid direct lexical overlap with the text of their ground-truth pages.

Figures

Figures reproduced from arXiv: 2606.23539 by Haocheng Wang, Tongkun Guan, Wei Shen, Xiaokang Yang.

Figure 1
Figure 1. Figure 1: Overview of the proposed method LightSTAR. Our method decomposes re￾trieval into LLM-Free Visual Selection and Vision-Adaptive Semantic Refinement, achieving state-of-the-art accuracy with lowest latency cost. To address this, we begin with a key empirical observation: in practical sce￾narios, user queries typically correspond to only a small subset of pages within a document collection. Because queries ar… view at source ↗
Figure 2
Figure 2. Figure 2: End-to-end retrieval latency comparison between LightSTAR with three com￾petitive MLLM-based retrievers (VisRAG-Ret, ColQwen2.5, and ColPali) as corpus size increases from 500 to 7,000 document pages. At 7,000 pages, LightSTAR is 10.4× faster than VisRAG-Ret, 4.1× faster than ColQwen2.5, and 2.3× faster than ColPali, demonstrating its computational efficiency for large-scale visual document retrieval. tive… view at source ↗
Figure 1
Figure 1. Figure 1: Ablation studies on key hyperparameters [PITH_FULL_IMAGE:figures/full_fig_p022_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Examples of token-level visual-text alignment. We visualize the similarity be￾tween query token samples and document images. Warmer colors indicate higher sim￾ilarity, showing that the visual encoder can localize textual regions associated with specific query tokens [PITH_FULL_IMAGE:figures/full_fig_p024_2.png] view at source ↗
read the original abstract

Visual document retrieval requires rapidly locating relevant pages from large multi-modal corpora in response to user queries. While recent methods powered by Multi-modal Large Language Models (MLLMs) show competitive accuracy, they suffer from prohibitive computational costs by applying intensive MLLM encoding to every single page. Meanwhile, we observe that user queries are typically keyword-anchored, containing semantically rich words that are expected to appear directly in the visible text of relevant pages, offering an efficient cue for quickly narrowing down candidate pages. Building on this insight, we propose LightSTAR, an efficient framework that decomposes visual document retrieval into: 1) LLM-free Visual Selection, which utilizes content-grounded query encoding to focus on informative words and employs LLM-free visual embeddings to produce a high-recall candidate set; and 2) Vision-adaptive Semantic Refinement, which further performs fine-grained semantic matching exclusively on these top candidates via adaptive region-wise feature fusion to effectively combine textual and layout cues, optimized through a hardness-aware contrastive objective. Experimental results demonstrate that LightSTAR achieves state-of-the-art retrieval accuracy while reducing end-to-end latency by several-fold, offering a highly practical solution to the accuracy-efficiency trade-off in visual document retrieval. Code is available at https://github.com/bokufa/LightSTAR.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes LightSTAR, a two-stage framework for visual document retrieval from large multi-modal corpora. It decomposes the task into (1) an LLM-free Visual Selection stage that uses content-grounded query encoding to identify informative words and produce a high-recall candidate set via visual embeddings, leveraging the observation that queries are typically keyword-anchored with direct lexical overlap in relevant page text, and (2) a Vision-adaptive Semantic Refinement stage that performs fine-grained matching on the top candidates using adaptive region-wise feature fusion of textual and layout cues, trained with a hardness-aware contrastive objective. Experiments claim state-of-the-art retrieval accuracy with several-fold end-to-end latency reduction compared to full MLLM baselines; code is released.

Significance. If the accuracy and latency claims are supported by the experiments, the work addresses a practical accuracy-efficiency trade-off in visual document retrieval by avoiding full-page MLLM encoding on all candidates. The code release supports reproducibility. The approach is grounded in an observable property of queries rather than purely learned parameters.

major comments (2)
  1. [Abstract, §3.1] Abstract and §3.1 (LLM-free Visual Selection): The high-recall guarantee of the candidate selection stage is load-bearing for the overall latency claim and rests on the premise that 'user queries are typically keyword-anchored, containing semantically rich words that are expected to appear directly in the visible text of relevant pages.' No quantitative analysis (e.g., fraction of queries exhibiting sufficient lexical overlap, performance breakdown on paraphrased vs. keyword queries, or coverage statistics on the evaluation datasets) is provided to establish how often this assumption holds; if it fails on a non-trivial subset, the refinement stage alone cannot recover SOTA accuracy without reintroducing full MLLM costs.
  2. [§4] §4 (Experiments): The manuscript reports SOTA accuracy and latency gains, but without access to the full experimental details, baselines, ablations, or error analysis in the provided text, it is not possible to verify whether the gains are robust or whether they depend on the keyword-anchored assumption holding in the test sets. A dedicated ablation or query-type breakdown is needed to substantiate the central claim.
minor comments (2)
  1. [§3.2] Notation for the adaptive region-wise fusion and hardness-aware contrastive loss should be introduced with explicit equations rather than prose descriptions to improve clarity.
  2. [Figures] Figure captions and axis labels in the latency-accuracy plots should explicitly state the datasets and metrics used.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the two major comments point by point below and will revise the manuscript to incorporate additional analysis as outlined.

read point-by-point responses

  1. Referee: [Abstract, §3.1] Abstract and §3.1 (LLM-free Visual Selection): The high-recall guarantee of the candidate selection stage is load-bearing for the overall latency claim and rests on the premise that 'user queries are typically keyword-anchored, containing semantically rich words that are expected to appear directly in the visible text of relevant pages.' No quantitative analysis (e.g., fraction of queries exhibiting sufficient lexical overlap, performance breakdown on paraphrased vs. keyword queries, or coverage statistics on the evaluation datasets) is provided to establish how often this assumption holds; if it fails on a non-trivial subset, the refinement stage alone cannot recover SOTA accuracy without reintroducing full MLLM costs.

    Authors: We agree that a quantitative characterization of the keyword-anchored assumption would strengthen the central claim. The current manuscript presents the observation as motivation but does not report explicit statistics on lexical overlap or query-type breakdowns. In the revision we will add a dedicated subsection (likely in §3 or §4) that measures (i) the fraction of queries exhibiting direct lexical overlap with relevant page text, (ii) coverage statistics on the evaluation datasets, and (iii) retrieval performance stratified by keyword-anchored versus paraphrased queries. This analysis will be performed on the same test sets used for the main results. revision: yes

  2. Referee: [§4] §4 (Experiments): The manuscript reports SOTA accuracy and latency gains, but without access to the full experimental details, baselines, ablations, or error analysis in the provided text, it is not possible to verify whether the gains are robust or whether they depend on the keyword-anchored assumption holding in the test sets. A dedicated ablation or query-type breakdown is needed to substantiate the central claim.

    Authors: The complete manuscript already contains the full experimental protocol, baselines, and main ablations; however, it does not yet include an explicit query-type breakdown. We will add this breakdown (keyword-anchored vs. paraphrased queries) together with an ablation that isolates the contribution of the selection stage under varying degrees of lexical overlap. These additions will directly address whether the reported accuracy and latency gains remain robust when the assumption holds only partially. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper describes an empirical framework grounded in an external observation about keyword-anchored queries, decomposed into LLM-free selection and vision-adaptive refinement stages, with performance claims supported by experimental results on retrieval accuracy and latency. No equations, self-referential definitions, fitted parameters renamed as predictions, or load-bearing self-citations are present in the provided text that would reduce any claimed result to its inputs by construction. The method is self-contained against external benchmarks via reported experiments.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework rests on one domain assumption about query structure; no free parameters or invented entities are mentioned in the abstract.

axioms (1)
  • domain assumption User queries are typically keyword-anchored, containing semantically rich words that are expected to appear directly in the visible text of relevant pages.
    This observation is presented as the foundation for the LLM-free selection stage.

pith-pipeline@v0.9.1-grok · 5765 in / 1185 out tokens · 23312 ms · 2026-06-26T08:41:31.651724+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 10 canonical work pages

  1. [1]

    In: Koyejo, S., Mohamed, S., Agarwal, A., Belgrave, D., Cho, K., Oh, A

    Alayrac, J.B., Donahue, J., Luc, P., Miech, A., Barr, I., Hasson, Y., Lenc, K., Men- sch, A., Millican, K., Reynolds, M., Ring, R., Rutherford, E., Cabi, S., Han, T., Gong, Z., Samangooei, S., Monteiro, M., Menick, J.L., Borgeaud, S., Brock, A., Nematzadeh, A., Sharifzadeh, S., Bińkowski, M.a., Barreira, R., Vinyals, O., Zisser- man, A., Simonyan, K.: Fla...

    2022

  2. [2]

    Bai, J., Bai, S., Yang, S., Wang, S., Tan, S., Wang, P., Lin, J., Zhou, C., Zhou, J.: Qwen-vl: A versatile vision-language model for understanding, localization, text reading, and beyond (2023),https://arxiv.org/abs/2308.12966

    Pith/arXiv arXiv 2023

  3. [3]

    Beyer, L., Steiner, A., Pinto, A.S., Kolesnikov, A., Wang, X., Salz, D., Neumann, M., Alabdulmohsin, I., Tschannen, M., Bugliarello, E., Unterthiner, T., Keysers, D., Koppula, S., Liu, F., Grycner, A., Gritsenko, A., Houlsby, N., Kumar, M., Rong, K., Eisenschlos, J., Kabra, R., Bauer, M., Bošnjak, M., Chen, X., Minderer, M., Voigtlaender, P., Bica, I., Ba...

    Pith/arXiv arXiv 2024

  4. [4]

    In: Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M., Lin, H

    Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J.D., Dhariwal, P., Nee- lakantan, A., Shyam, P., Sastry, G., Askell, A., Agarwal, S., Herbert-Voss, A., Krueger, G., Henighan, T., Child, R., Ramesh, A., Ziegler, D., Wu, J., Winter, C., Hesse, C., Chen, M., Sigler, E., Litwin, M., Gray, S., Chess, B., Clark, J., Berner, C., McCandlish, S., Radford, A....

    1901

  5. [5]

    M3- Embedding : Multi - Linguality , Multi - Functionality , Multi - Granularity Text Embeddings Through Self - Knowledge Distillation

    Chen, J., Xiao, S., Zhang, P., Luo, K., Lian, D., Liu, Z.: M3-embedding: Multi- linguality, multi-functionality, multi-granularity text embeddings through self- knowledge distillation. In: Ku, L.W., Martins, A., Srikumar, V. (eds.) Findings of the Association for Computational Linguistics: ACL 2024. pp. 2318–2335. As- sociation for Computational Linguisti...

    work page doi:10.18653/v1/2024.findings-acl.137 2024

  6. [6]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Chen, Z., Wu, J., Wang, W., Su, W., Chen, G., Xing, S., Zhong, M., Zhang, Q., Zhu, X., Lu, L., et al.: Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 24185–24198 (2024)

    2024

  7. [7]

    In: Koyejo, S., Mohamed, S., Agar- wal, A., Belgrave, D., Cho, K., Oh, A

    Dao, T., Fu, D., Ermon, S., Rudra, A., Ré, C.: Flashattention: Fast and memory- efficient exact attention with io-awareness. In: Koyejo, S., Mohamed, S., Agar- wal, A., Belgrave, D., Cho, K., Oh, A. (eds.) Advances in Neural Informa- tion Processing Systems. vol. 35, pp. 16344–16359. Curran Associates, Inc. (2022),https : / / proceedings . neurips . cc / ...

    2022

  8. [8]

    Guan et al

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., Houlsby, N.: 16 T. Guan et al. An image is worth 16x16 words: Transformers for image recognition at scale. In: In- ternational Conference on Learning Representations (2021),https://openreview. net/forum...

    2021

  9. [9]

    In: Yue, Y., Garg, A., Peng, N., Sha, F., Yu, R

    Faysse, M., Sibille, H., Wu, T., Omrani, B., Viaud, G., HUDELOT, C., Colombo, P.: Colpali: Efficient document retrieval with vision language models. In: Yue, Y., Garg, A., Peng, N., Sha, F., Yu, R. (eds.) International Conference on Learn- ing Representations. vol. 2025, pp. 61424–61449 (2025),https://proceedings. iclr.cc/paper_files/paper/2025/file/99e9e...

    2025

  10. [10]

    In: Che, W., Nabende, J., Shutova, E., Pilehvar, M.T

    Fu, P., Guan, T., Wang, Z., Guo, Z., Duan, C., Sun, H., Chen, B., Jiang, Q., Ma, J., Zhou, K., Luo, J.: Multimodal large language models for text-rich image understanding: A comprehensive review. In: Che, W., Nabende, J., Shutova, E., Pilehvar, M.T. (eds.) Findings of the Association for Computational Linguistics: ACL 2025. pp. 19941–19958 (2025)

    2025

  11. [11]

    arXiv preprint arXiv:2410.16261 (2024)

    Gao, Z., Chen, Z., Cui, E., Ren, Y., Wang, W., Zhu, J., Tian, H., Ye, S., He, J., Zhu, X., et al.: Mini-internvl: A flexible-transfer pocket multimodal model with 5% parameters and 90% performance. arXiv preprint arXiv:2410.16261 (2024)

    arXiv 2024

  12. [12]

    IEEE Transactions on Circuits and Systems for Video Technology32(9), 6073–6085 (2022)

    Guan, T., Gu, C., Lu, C., Tu, J., Feng, Q., Wu, K., Guan, X.: Industrial scene text detection with refined feature-attentive network. IEEE Transactions on Circuits and Systems for Video Technology32(9), 6073–6085 (2022)

    2022

  13. [13]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Guan, T., Gu, C., Tu, J., Yang, X., Feng, Q., Zhao, Y., Shen, W.: Self-supervised implicit glyph attention for text recognition. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 15285– 15294 (June 2023)

    2023

  14. [14]

    In: CVPR

    Guan, T., Gu, C., Tu, J., Yang, X., Feng, Q., Zhao, Y., Shen, W.: Self-supervised implicit glyph attention for text recognition. In: CVPR. pp. 15285–15294 (2023)

    2023

  15. [15]

    Guan,T.,Lin,C.,Shen,W.,Yang,X.:Posformer:recognizingcomplexhandwritten mathematicalexpressionwithpositionforesttransformer.In:EuropeanConference on Computer Vision. pp. 130–147. Springer (2025)

    2025

  16. [16]

    IEEE Transactions on Pattern Analysis and Machine Intelligence (2025)

    Guan, T., Shen, W., Yang, X.: CCDPlus: Towards accurate character to charac- ter distillation for text recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence (2025)

    2025

  17. [17]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Guan, T., Shen, W., Yang, X., Feng, Q., Jiang, Z., Yang, X.: Self-supervised character-to-character distillation for text recognition. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 19473–19484 (2023)

    2023

  18. [18]

    In: European Conference on Computer Vision

    Guan, T., Shen, W., Yang, X., Wang, X., Yang, X.: Bridging synthetic and real worlds for pre-training scene text detectors. In: European Conference on Computer Vision. pp. 428–446. Springer (2024)

    2024

  19. [19]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Guan, T., Wang, Z., Fu, P., Guo, Z., Shen, W., Zhou, K., Yue, T., Duan, C., Sun, H., Jiang, Q., et al.: A token-level text image foundation model for docu- ment understanding. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 23210–23220 (2025)

    2025

  20. [20]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Guan, T., Yang, Z., Wan, J., Yang, M., Guo, Z., Hu, Z., Luo, R., Chen, R., Jiang, S., Wang, P., et al.: Codepercept: Code-grounded visual stem perception for mllms. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 33542–33552 (2026)

    2026

  21. [21]

    In: Adelani, D.I., Ar- nett, C., Ataman, D., Chang, T.A., Gonen, H., Raja, R., Schmidt, F., Stap, D., LightSTAR for Efficient Visual Document Retrieval 17 Wang, J

    Günther, M., Sturua, S., Akram, M.K., Mohr, I., Ungureanu, A., Wang, B., Es- lami, S., Martens, S., Werk, M., Wang, N., Xiao, H.: jina-embeddings-v4: Uni- versal embeddings for multimodal multilingual retrieval. In: Adelani, D.I., Ar- nett, C., Ataman, D., Chang, T.A., Gonen, H., Raja, R., Schmidt, F., Stap, D., LightSTAR for Efficient Visual Document Ret...

    work page doi:10.18653/v1/2025.mrl-main.36 2025

  22. [22]

    In: Proceed- ings of the 32nd International ACM SIGIR Conference on Research and Develop- ment in Information Retrieval

    Guo, J., Xu, G., Cheng, X., Li, H.: Named entity recognition in query. In: Proceed- ings of the 32nd International ACM SIGIR Conference on Research and Develop- ment in Information Retrieval. p. 267–274. SIGIR ’09, Association for Computing Machinery, New York, NY, USA (2009).https://doi.org/10.1145/1571941. 1571989,https://doi.org/10.1145/1571941.1571989

    work page doi:10.1145/1571941 2009

  23. [23]

    In: International Con- ference on Learning Representations (2022),https://openreview.net/forum?id= nZeVKeeFYf9

    Hu, E.J., yelong shen, Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., Chen, W.: LoRA: Low-rank adaptation of large language models. In: International Con- ference on Learning Representations (2022),https://openreview.net/forum?id= nZeVKeeFYf9

    2022

  24. [24]

    Jiang, T., Song, M., Zhang, Z., Huang, H., Deng, W., Sun, F., Zhang, Q., Wang, D., Zhuang, F.: E5-v: Universal embeddings with multimodal large language models (2024),https://arxiv.org/abs/2407.12580

    Pith/arXiv arXiv 2024

  25. [25]

    In: The Thirteenth International Conference on Learning Representations (2025),https: //openreview.net/forum?id=TE0KOzWYAF

    Jiang, Z., Meng, R., Yang, X., Yavuz, S., Zhou, Y., Chen, W.: VLM2vec: Train- ing vision-language models for massive multimodal embedding tasks. In: The Thirteenth International Conference on Learning Representations (2025),https: //openreview.net/forum?id=TE0KOzWYAF

    2025

  26. [26]

    Kaplan, J., McCandlish, S., Henighan, T., Brown, T.B., Chess, B., Child, R., Gray, S., Radford, A., Wu, J., Amodei, D.: Scaling laws for neural language models (2020),https://arxiv.org/abs/2001.08361

    Pith/arXiv arXiv 2020

  27. [27]

    Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =

    Khattab, O., Zaharia, M.: Colbert: Efficient and effective passage search via con- textualized late interaction over BERT. In: Huang, J.X., Chang, Y., Cheng, X., Kamps, J., Murdock, V., Wen, J., Liu, Y. (eds.) Proceedings of the 43rd Inter- national ACM SIGIR conference on research and development in Information Re- trieval, SIGIR 2020, Virtual Event, Chi...

    work page doi:10.1145/3397271.3401075 2020

  28. [28]

    In: Computer Vision – ECCV 2022

    Kim, G., Hong, T., Yim, M., Nam, J., Park, J., Yim, J., Hwang, W., Yun, S., Han, D., Park, S.: Ocr-free document understanding transformer. In: Computer Vision – ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part XXVIII. p. 498–517. Springer-Verlag, Berlin, Heidelberg (2022). https://doi.org/10.1007/978-3-031-198...

    work page doi:10.1007/978-3-031-19815-1_29 2022

  29. [29]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y., et al.: Segment anything. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 4015–4026 (2023)

    2023

  30. [30]

    In: Krause, A., Brunskill, E., Cho, K., Engelhardt, B., Sabato, S., Scarlett, J

    Lee, K., Joshi, M., Turc, I.R., Hu, H., Liu, F., Eisenschlos, J.M., Khandelwal, U., Shaw, P., Chang, M.W., Toutanova, K.: Pix2Struct: Screenshot parsing as pre- training for visual language understanding. In: Krause, A., Brunskill, E., Cho, K., Engelhardt, B., Sabato, S., Scarlett, J. (eds.) Proceedings of the 40th Inter- national Conference on Machine Le...

    2023

  31. [31]

    In: Larochelle, H., Ran- zato, M., Hadsell, R., Balcan, M., Lin, H

    Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küt- tler, H., Lewis, M., Yih, W.t., Rocktäschel, T., Riedel, S., Kiela, D.: Retrieval- augmented generation for knowledge-intensive nlp tasks. In: Larochelle, H., Ran- zato, M., Hadsell, R., Balcan, M., Lin, H. (eds.) Advances in Neural Infor- 18 T. Guan et al. mation Processing Sys...

    2020

  32. [32]

    In: Chaudhuri, K., Jegelka, S., Song, L., Szepesvari, C., Niu, G., Sabato, S

    Li, J., Li, D., Xiong, C., Hoi, S.: BLIP: Bootstrapping language-image pre- training for unified vision-language understanding and generation. In: Chaudhuri, K., Jegelka, S., Song, L., Szepesvari, C., Niu, G., Sabato, S. (eds.) Proceedings of the 39th International Conference on Machine Learning. Proceedings of Ma- chine Learning Research, vol. 162, pp. 1...

    2022

  33. [33]

    In: Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., Levine, S

    Liu, H., Li, C., Wu, Q., Lee, Y.J.: Visual instruction tuning. In: Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., Levine, S. (eds.) Advances in Neu- ral Information Processing Systems. vol. 36, pp. 34892–34916. Curran Associates, Inc. (2023),https://proceedings.neurips.cc/paper_files/paper/2023/file/ 6dcf277ea32ce3288914faf369fe6de0-Paper-Conf...

    2023

  34. [34]

    org/abs/cs/0205028

    Loper, E., Bird, S.: Nltk: The natural language toolkit (2002),https://arxiv. org/abs/cs/0205028

    Pith/arXiv arXiv 2002

  35. [35]

    doi:10.18653/v1/2023.eacl- main.240

    Muennighoff, N., Tazi, N., Magne, L., Reimers, N.: MTEB: Massive text embed- ding benchmark. In: Vlachos, A., Augenstein, I. (eds.) Proceedings of the 17th Conference of the European Chapter of the Association for Computational Lin- guistics. pp. 2014–2037. Association for Computational Linguistics, Dubrovnik, Croatia (May 2023).https://doi.org/10.18653/v...

    work page doi:10.18653/v1/2023.eacl- 2014

  36. [36]

    Nussbaum, Z., Duderstadt, B., Mulyar, A.: Nomic embed vision: Expanding the latent space (2024),https://arxiv.org/abs/2406.18587

    arXiv 2024

  37. [37]

    van den Oord, A., Li, Y., Vinyals, O.: Representation learning with contrastive predictive coding (2019),https://arxiv.org/abs/1807.03748

    Pith/arXiv arXiv 2019

  38. [38]

    In: Meila, M., Zhang, T

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., Sutskever, I.: Learning transferable visual models from natural language supervision. In: Meila, M., Zhang, T. (eds.) Proceedings of the 38th International Conference on Machine Learning. Proceed- ings of Machine Learning Res...

    2021

  39. [39]

    Robertson, S., Zaragoza, H.: The probabilistic relevance framework: Bm25 and beyond. Found. Trends Inf. Retr.3(4), 333–389 (Apr 2009).https://doi.org/ 10.1561/1500000019,https://doi.org/10.1561/1500000019

    work page doi:10.1561/1500000019 2009

  40. [40]

    Robertson,S.E.,Walker,S.,Jones,S.,Hancock-Beaulieu,M.M.,Gatford,M.,etal.: Okapi at trec (1994)

    1994

  41. [41]

    arXiv preprint arXiv:2508.10104 (2025)

    Siméoni, O., Vo, H.V., Seitzer, M., Baldassarre, F., Oquab, M., Jose, C., Khali- dov, V., Szafraniec, M., Yi, S., Ramamonjisoa, M., et al.: Dinov3. arXiv preprint arXiv:2508.10104 (2025)

    Pith/arXiv arXiv 2025

  42. [42]

    Journal of Documentation , volume =

    SPARCK JONES, K.: A statistical interpretation of term specificity and its ap- plication in retrieval. Journal of Documentation28(1), 11–21 (01 1972).https: //doi.org/10.1108/eb026526,https://doi.org/10.1108/eb026526

    work page doi:10.1108/eb026526 1972

  43. [43]

    Team, N.: Nomic embed multimodal: Interleaved text, image, and screenshots for visual document retrieval (2025),https://nomic.ai/blog/posts/nomic-embed- multimodal

    2025

  44. [44]

    Thakur, N., Reimers, N., Rücklé, A., Srivastava, A., Gurevych, I.: BEIR: A het- erogeneous benchmark for zero-shot evaluation of information retrieval models. In: Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2) (2021),https://openreview.net/forum?id= wCu6T5xFjeJ LightSTAR for Efficient Visual Docume...

    2021

  45. [45]

    arXiv preprint arXiv:2502.14786 (2025)

    Tschannen, M., Gritsenko, A., Wang, X., Naeem, M.F., Alabdulmohsin, I., Parthasarathy, N., Evans, T., Beyer, L., Xia, Y., Mustafa, B., et al.: Siglip 2: Multilingual vision-language encoders with improved semantic understanding, lo- calization, and dense features. arXiv preprint arXiv:2502.14786 (2025)

    Pith/arXiv arXiv 2025

  46. [46]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Wang,Z.,Guan,T.,Fu,P.,Duan,C.,Jiang,Q.,Guo,Z.,Guo,S.,Luo,J.,Shen,W., Yang, X.: Marten: Visual question answering with mask generation for multi-modal document understanding. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 14460–14471 (2025)

    2025

  47. [47]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Wen, C., Peng, Z., Huang, Y., Shen, W.: Efficient segmentation with multimodal large language model via token routing. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 40, pp. 10593–10602 (2026)

    2026

  48. [48]

    Prefix-Tuning: Optimizing Continuous Prompts for Generation

    Xu, Y., Xu, Y., Lv, T., Cui, L., Wei, F., Wang, G., Lu, Y., Florencio, D., Zhang, C., Che, W., Zhang, M., Zhou, L.: LayoutLMv2: Multi-modal pre-training for visually- rich document understanding. In: Zong, C., Xia, F., Li, W., Navigli, R. (eds.) Pro- ceedings of the 59th Annual Meeting of the Association for Computational Linguis- tics and the 11th Intern...

    work page doi:10.18653/v1/2021.acl- 2021

  49. [49]

    In: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining(KDD).pp.1866–1876.ACM(2020).https://doi.org/10.1145/3394486

    Xu,Y.,Li,M.,Cui,L.,Huang,S.,Wei,F.,Zhou,M.:Layoutlm:Pre-trainingoftext and layout for document image understanding. In: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. p. 1192–1200. KDD ’20, ACM (Aug 2020).https://doi.org/10.1145/3394486. 3403172,http://dx.doi.org/10.1145/3394486.3403172

    work page doi:10.1145/3394486 2020

  50. [50]

    In: Yue, Y., Garg, A., Peng, N., Sha, F., Yu, R

    Yu, S., Tang, C., Xu, B., Cui, J., Ran, J., Yan, Y., Liu, Z., Wang, S., Han, X., Liu, Z., Sun, M.: Visrag: Vision-based retrieval-augmented generation on multi-modality documents. In: Yue, Y., Garg, A., Peng, N., Sha, F., Yu, R. (eds.) International Conference on Learning Representations. vol. 2025, pp. 21074– 21098 (2025),https://proceedings.iclr.cc/pape...

    2025

  51. [51]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    Zhai, X., Mustafa, B., Kolesnikov, A., Beyer, L.: Sigmoid loss for language im- age pre-training. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 11975–11986 (October 2023)

    2023

  52. [52]

    Zhang, J., Zhang, Q., Wang, B., Ouyang, L., Wen, Z., Li, Y., Chow, K.H., He, C., Zhang, W.: Ocr hinders rag: Evaluating the cascading impact of ocr on retrieval- augmented generation (2025),https://arxiv.org/abs/2412.02592

    arXiv 2025

  53. [53]

    what percentage of onlineusers watch cute animal clips?

    Zhu, J., Wang, W., Chen, Z., Liu, Z., Ye, S., Gu, L., Tian, H., Duan, Y., Su, W., Shao, J., Gao, Z., Cui, E., Wang, X., Cao, Y., Liu, Y., Wei, X., Zhang, H., Wang, H., Xu, W., Li, H., Wang, J., Deng, N., Li, S., He, Y., Jiang, T., Luo, J., Wang, Y., He, C., Shi, B., Zhang, X., Shao, W., He, J., Xiong, Y., Qu, W., Sun, P., Jiao, P., Lv, H., Wu, L., Zhang, ...

    Pith/arXiv arXiv 2025