An archaeological Catalog Collection Method Based on Large Vision-Language Models

Honglin Pang; Tianjing Duan; Xi Yang; Yi Chang

REVIEW 4 major objections 6 minor 18 references

An archaeological Catalog Collection Method Based on Large Vision-Language Models

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A three-module pipeline—document localization, block comprehension, block matching—lets large vision-language models extract pottery catalog entries, reaching 35.4% average precision, 33.8 points above the best direct baseline.

desk verdict The pipeline is a sensible engineering combination, but the evaluation's ground truth is the method's own corrected output, so the reported AP numbers don't support the reliability claim. read the letter →

arxiv 2412.20088 v1 pith:UHSARIIK submitted 2024-12-28 cs.CV cs.AI

Honglin Pang , Yi Chang , Tianjing Duan , Xi Yang This is my paper · ORCID

classification cs.CVcs.AI

keywords archaeologicalcatalogvision-languagemodelsopen-setobjectdetectiondocumentlocalizationblockmatchingmodalpotteryculturalheritagedigitization

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Archaeological catalogs—artifact images paired with excavation unit, class, and measurements—are scattered across thousands of PDFs, and existing vision-language models either cannot detect small artifact images or cannot link them to their text rows. This paper claims that a three-module pipeline solves both problems: an open-set detector isolates image and text blocks, a vision-language model converts each block into structured attributes, and a matcher aligns images to text by shared catalog numbers, using spatial distance to break ties. On 302 pages from two Neolithic pottery catalogs, the pipeline with the strongest tested model reaches 35.4% average precision at 90% IoU, a 33.8-point improvement over the best direct VLM baseline. If the result holds, automated collection of archaeological catalogs becomes practical, turning scattered legacy publications into structured data for classification, dating, and evolution studies.

What carries the argument

The load-bearing object is the matching module operating on structured attribute blocks. Each detected block is converted by a VLM into a dictionary of attributes; foreign-key matching uses the Jaccard overlap of the key sets, accepting a pair only when the overlap is exactly 1, and the distance-bipartite stage resolves the resulting many-to-many groups by solving a minimum-weight assignment on Euclidean distances between bounding-box centers. The design converts modal alignment from an open-ended grounding problem into two closed-form matching problems, which is what lets a general-purpose VLM succeed without specialized detection heads or fine-tuning.

What would settle it

Take a random sample of the 302 catalog pages, have two annotators who have never seen the pipeline's outputs transcribe all image–caption pairs by hand, and recompute the pipeline's AP against that independent ground truth; if the number approaches the direct-baseline level (1.6% AP) or the human annotations diverge substantially from the paper's cleaned data, the central reliability claim is false.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that direct VLM prompting fails at catalog collection because a single model cannot reliably localize small artifact images and bind them to their caption rows, while splitting the task into three modular steps succeeds. An open-set detector segments a page into image and text blocks; a VLM converts each block into a structured attribute dictionary (index, item code, class); and a matcher aligns the two modalities by accepting pairs whose foreign-key sets overlap completely, then resolves many-to-many groups with minimum-weight bipartite matching on the Euclidean distance between block centers. With Claude 3.5 Sonnet driving comprehension, the full pipeline reaches 35.4% AP on the cleaned ground truth, versus 1.6% AP for Qwen-VL used alone and 0% AP for a detector-plus-VLM combination without the matching module; GPT-4o inside the pipeline reaches 32.8% AP. These margins are the paper's evidence that all three modules are necessary and that the pipeline transfers across different vision-language models.

Load-bearing premise

The evaluation assumes that the cleaned dataset used as ground truth is an independent gold standard, even though the dataset was assembled by the proposed method with minimal manual correction, so any systematic failures that the corrections did not remove would inflate the measured AP.

Editorial extensions

If this is right

Automated collection of archaeological catalog pages is feasible with current off-the-shelf VLMs when localization and matching are handled by dedicated modules, rather than by the VLM alone.
The pipeline is model-agnostic but backend-sensitive: swapping the comprehension VLM moves AP between 27.4% and 35.4%.
The 90% IoU threshold means the gains are in precise pixel-level localization of artifact images, not just coarse page-level understanding.
The resulting 2,301-pair dataset with 44 shape classes and 310 excavation units provides a structured resource for downstream archaeological tasks such as classification and dating.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The reported 35.4% AP is best read as an upper bound because the ground truth came from the method itself with manual correction; a fresh independent annotation of a page subset would give the realistic number.
The method's matching rules depend on the regular layout of pottery catalogs, where an index and item code sit beside each image; irregular archaeological documents would likely need learned matching rather than fixed foreign keys.
A straightforward stress test is to rotate pages, merge columns, or erase item codes and measure how much of the 33.8-point gain survives, which would separate the contribution of layout regularity from the VLM's semantic comprehension.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The short version: this paper assembles three well-known pieces (open-set detection, VLM comprehension, foreign-key plus bipartite matching) into a pipeline for extracting pottery catalog entries from PDF pages. That assembly is reasonable and the problem is real. But the evaluation is circular, so the headline numbers — 35.4% AP with Claude 3.5 Sonnet, a 33.8-point jump over Qwen-VL alone — tell you more about the cleaning process than about the method.

The design is clean and the matching stage is a sensible answer to a genuine many-to-many problem in catalogs. The authors correctly identify that off-the-shelf VLMs can't ground artifact images reliably, and they show a non-trivial baseline comparison. The use of AP at IoU 0.9 is strict and appropriate.

The load-bearing problem is in Sections 4.1–4.2. The ground truth is the 'cleaned dataset' produced by the proposed method with 'minimal manual correction.' That makes the AP a self-consistency score. We don't know how much the human corrected, what the protocol was, or whether the corrections were guided by the method's failures. Without an independent annotation set, the improvement over baselines is not evidence of external accuracy. The paper also leaves GPT-4o and Claude-only baselines as '-' without explaining what that means, provides no error bars, no released code/data, and only two catalogs. Minor: Figure 4's text appears garbled in the PDF, but that's production not substance.

If the evaluation were fixed — e.g., a separate human annotation produced without seeing the method's output, plus a description of the correction load — the result could be useful to the digital heritage community. As it stands, the abstract's claim of a 'reliable solution' is not supported.

I would not send this to peer review in its current form; the circularity is load-bearing. I'd desk reject with an invitation to resubmit after a proper independent evaluation. For a reading group, it's a useful case study in how evaluation design can invalidate an otherwise coherent pipeline.

Referee Report

4 major / 6 minor

Summary. The paper proposes a three-module pipeline for automated archaeological catalog collection from scanned pottery catalog PDFs: a document localization module based on an open-set object detection model, a block comprehension module that uses a vision-language model (VLM) to convert image and text blocks into structured attribute dictionaries, and a block matching module that aligns image and text blocks through foreign-key overlap followed by distance-based bipartite matching. The authors report experiments on 302 pages from the Dabagou and Miaozigou pottery catalogs, claiming an AP of 35.4% with Claude 3.5 Sonnet, a 33.8-percentage-point improvement over Qwen-VL alone, and conclude that the approach provides a reliable solution for automated archaeological catalog collection.

Significance. If the claimed accuracy were established, the pipeline would be a practically useful contribution to cultural-heritage digitization and to VLM-based document understanding, since it couples open-set detection with structured attribute extraction and a principled matching rule. The three-module decomposition is sensible, and the inclusion of a no-pipeline baseline is a reasonable experimental design. However, the significance is currently capped by the circularity of the evaluation, the incomplete baseline table, and the absence of released data or code; these issues must be resolved before the effectiveness and reliability claims can be accepted.

major comments (4)

[Sections 4.1 and 4.2] The evaluation is circular: Section 4.1 states that the dataset was collected "using the proposed method" with "minimal manual correction," and Section 4.2 states "We use the cleaned dataset as ground truth to calculate metrics." The ground-truth boxes, class labels, and image-text pairings used to compute AP are therefore derived from the same detection, comprehension, and matching pipeline being evaluated. The paper does not quantify the manual corrections (e.g., how many boxes were added, deleted, or moved; how many class labels were corrected; how many pairings were fixed) and does not describe an annotation protocol that would prevent the model's outputs from biasing the corrections. This is load-bearing because the abstract and Section 5 base the "reliable solution" and "significantly improves accuracy" claims on the AP values in Table 1. The authors should construct an independent gold standard, for example by having annotators label pages without seeing the method's outputs, and report the comparison on that gold standard together with correction statistics.
[Table 1 and Section 4.2] The baseline comparison is incomplete: "GPT-4o only" and "Claude 3.5 Sonnet only" are listed as "-" without any score or explicit failure criterion, and the statement that these models "lack object detection capabilities" is an assertion rather than a measured result. The headline 33.8-point improvement is computed only against Qwen-VL-only, so the reader cannot verify that the pipeline outperforms all direct VLM baselines. In addition, all AP values appear to come from a single evaluation without error bars, confidence intervals, or significance tests; given the strongly long-tailed class distribution shown in Figure 4(a), the reported differences may be driven by a few frequent classes. Please report failure rates for the incomplete baselines, per-class or at least page-level bootstrap intervals, and the exact detection recall before matching.
[Section 3.2, Eq. (1); Section 3.4, Eq. (4); Section 4.2] The detection model used inside the proposed pipeline is never identified: Eq. (1) refers only to "open-set object detection model," while the baseline explicitly uses Grounding-DINO. The reader cannot tell whether the proposed method also uses Grounding-DINO or a different detector, which makes the method irreproducible and leaves open the possibility that the Table 1 gain comes from detector choice rather than from the proposed pipeline. Please specify the detection model, the exact prompts P_img, P_text, and P_c, the VLM versions and decoding settings, and the concrete foreign-key attribute set K_foreign used in Eq. (4).
[Section 4.2 and Eqs. (5)-(9)] Two thresholds that materially affect the reported numbers are asserted without justification or sensitivity analysis: the foreign-key matching threshold M = 1 in Eq. (6) and the AP IoU threshold of 0.9 in Section 4.2. A single OCR or comprehension error in a foreign key makes M < 1 and rejects a correct pair, and an IoU threshold of 0.9 is unusually strict for detection evaluation. Please report AP at standard IoU thresholds (e.g., 0.5 and 0.75), vary the matching threshold (e.g., 0.75, 0.9, 1.0), and justify the chosen operating points with data.

minor comments (6)

[Sections 4.1 and 4.2] The manuscript contains two objects labeled "Table 1": one for artifact-per-unit statistics in Section 4.1 and one for the performance comparison in Section 4.2. Renumber the tables to avoid ambiguity.
[Figure 4] The axis labels in Figure 4 are rendered as garbled Unicode escape sequences (e.g., "/uni00000026/uni00000044/...") rather than readable category names; replace them with the actual class and unit labels.
[ACM reference block] The ACM reference block contains placeholder text ("Make sure to enter the correct conference title from your rights confirmation email") and a 2018 copyright date for a 2025 submission; these must be corrected before publication.
[Eq. (8)] The notation is inconsistent: Eq. (8) writes M* in the minimization formula while the surrounding text refers to M*_k. Make the notation uniform.
[Section 3.4] The foreign-key attribute set K_foreign is used in Eq. (4) but never explicitly defined; give a concrete example for the pottery catalogs (e.g., which attribute keys are treated as foreign keys).
[Section 4.1] The term "data pair" is not formally defined; clarify whether it denotes one image-text correspondence, one artifact record, or one page-level pair.

Circularity Check

1 steps flagged · score 6.0 of 10

Evaluation is circular: the 'cleaned dataset' used as ground truth in Section 4.2 was built by the proposed method itself in Section 4.1, so the AP gains in Table 1 do not independently validate the claimed reliable solution.

self definitional [Section 4.1 (Data Collection and Statistics) and Section 4.2 (Comparison Experiments)]
"Using the proposed method, we collected 302 pottery catalog PDF pages from Dabagou and Miaozigou sites. With minimal manual correction, we obtained 2,301 data pairs, demonstrating the effectiveness of our approach. / We use the cleaned dataset as ground truth to calculate metrics for results obtained from each method, in order to evaluate their effectiveness."

The ground truth for every metric is the 'cleaned dataset' that Section 4.1 says was produced 'using the proposed method' with only 'minimal manual correction'. The reference bounding boxes, class labels, and image-text pairings used for AP at IoU 90% are therefore outputs of the same detection and comprehension modules being scored, not an independently established gold standard. The paper gives no annotation protocol, no correction statistics, and no external benchmark, so the reported 33.8-point improvement over Qwen-VL alone largely measures agreement with a lightly corrected version of the method's own output. This dependency is load-bearing because all comparison experiments and the conclusion of reliability rest on Table 1.

full rationale

The paper's pipeline itself is internally coherent: document localization, block comprehension, and bipartite block matching are described as concrete modules, and the matching equations (4)-(9) do not presuppose the evaluation result. The self-citations in the reference list ([9], [12], [16]) are background and related-work citations and are not used to license the central claim. However, the central claim of a 'reliable solution' is validated exclusively against a ground-truth dataset that Section 4.1 states was obtained 'using the proposed method' with 'minimal manual correction'. Section 4.2 then treats that same cleaned dataset as the ground truth for AP computation. Unless the unspecified manual corrections were a full independent re-annotation, the metric is circular: the system is scored against a cleaned version of its own outputs. If the corrections were extensive, the paper's characterization is misleading and the difficulty of the task is hidden; if they were minimal, the AP values mostly demonstrate self-consistency rather than external accuracy. The absence of released code, data, or annotation protocol makes the dependency uncheckable. This affects the main effectiveness claim and the 33.8% improvement, so the circularity score is 6 rather than a lower score; it is partial because the pipeline architecture and the baseline comparison design have independent content.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The pipeline relies on standard pretrained models and classic matching algorithms. The main design choices are the foreign key attribute set, the strict matching threshold, and the evaluation threshold. The most consequential premise is the independence of the cleaned ground truth, which is not established.

free parameters (3)

foreign key attribute set K_foreign
The attributes used for matching (e.g., index/item number) are selected by the authors and directly determine whether foreign key matching succeeds; not fitted to data but a hand-chosen design choice that gates the central result.
matching threshold M = 1 = 1
Equation (5)-(6) require complete overlap of foreign key sets for a match; this binary threshold is chosen without justification and ignores partial overlaps.
IoU threshold 0.9 for AP evaluation = 0.9
The evaluation uses a strict IoU threshold of 90%, chosen by the authors; changing this threshold would change all reported AP values, but this is an evaluation parameter rather than a model parameter.

assumptions (3)

domain assumption The archaeological catalogs contain structured blocks linked by foreign keys such as artifact index numbers.
The entire matching module (Sec. 3.4) assumes that each image and text block carries a common index or item attribute that can be extracted reliably; if the catalogs lack such keys, foreign key matching fails.
domain assumption Open-set object detection with prompt 'Objects.Words' reliably segments image and text blocks in catalog pages.
The Document Localization Module (Sec. 3.2) depends on the detection model's ability to localize all relevant blocks; no detection accuracy is reported separately.
ad hoc to paper The manually corrected dataset is an unbiased ground truth.
Section 4.2 uses the cleaned dataset as ground truth without describing the correction process; the independence of this ground truth from the method is asserted, not established.

how reviews work

0 comments

Cite this review

Pith. "Pith review of An archaeological Catalog Collection Method Based on Large Vision-Language Models." pith.science (2026). https://pith.science/paper/UHSARIIK

@misc{pith2026241220088,
  author       = {Pith},
  title        = {Pith review of: An archaeological Catalog Collection Method Based on Large Vision-Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UHSARIIK}},
  note         = {Machine review of arXiv:2412.20088}
}

read the original abstract

Archaeological catalogs, containing key elements such as artifact images, morphological descriptions, and excavation information, are essential for studying artifact evolution and cultural inheritance. These data are widely scattered across publications, requiring automated collection methods. However, existing Large Vision-Language Models (VLMs) and their derivative data collection methods face challenges in accurate image detection and modal matching when processing archaeological catalogs, making automated collection difficult. To address these issues, we propose a novel archaeological catalog collection method based on Large Vision-Language Models that follows an approach comprising three modules: document localization, block comprehension and block matching. Through practical data collection from the Dabagou and Miaozigou pottery catalogs and comparison experiments, we demonstrate the effectiveness of our approach, providing a reliable solution for automated collection of archaeological catalogs.

Figures

Figures reproduced from arXiv: 2412.20088 by the authors.

**Figure 1.** Target tasks for catalog datasets. datasets can be leveraged for various computational tasks, including classification tasks for artifact categorization, generative tasks for artifacts reconstruction, and visual question answering tasks, as shown in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗

**Figure 2.** The pipeline consists of three main modules: Document Localization Module for segmenting image and text regions, [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** Annotations of a pottery example. where 𝑘𝑗 represents attribute keys and 𝑣𝑗 represents corresponding values. 3.4 Block Matching Module To establish correspondences between image blocks and text blocks, we propose a two-stage matching strategy. Foreign Key Matching. For each element 𝑐𝑞 in image region comprehension results 𝐶𝑖𝑚𝑔 and text region comprehension results 𝐶𝑡𝑒𝑥𝑡, we define its foreign key set 𝐹𝐾𝑞 as: 𝐹𝐾𝑞 = {… view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Statistical analysis of the dataset: (a) shows the distribution of samples across different categories, and (b) illustrates [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

18 extracted references · 7 canonical work pages

[1]

Anthropic. 2024. Claude 3.5 Sonnet Model Card Addendum. https://www- cdn.anthropic.com/fed9cc193a14b84131812372d8d5857f8f304c52/Model_Card_ Claude_3_Addendum.pdf

2024
[2]

Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. 2023. Qwen-vl: A frontier large vision- language model with versatile abilities. arXiv preprint arXiv:2308.12966 (2023)

arXiv 2023
[3]

Anjanava Biswas and Wrick Talukdar. 2024. Robustness of Structured Data Extraction from In-Plane Rotated Documents Using Multi-Modal Large Language Models (LLM). Journal of Artificial Intelligence Research (2024)

work page 2024
[4]

Xiang Chen, Chenxi Wang, Yida Xue, Ningyu Zhang, Xiaoyan Yang, Qiang Li, Yue Shen, Lei Liang, Jinjie Gu, and Huajun Chen. 2024. Unified hallucination detection for multimodal large language models. arXiv preprint arXiv:2402.03190 (2024)

arXiv 2024
[5]

Wentao He, Hanjie Ma, Shaohua Li, Hui Dong, Haixiang Zhang, and Jie Feng

work page
[6]

Liunian Harold Li, Pengchuan Zhang, Haotian Zhang, Jianwei Yang, Chunyuan Li, Yiwu Zhong, Lijuan Wang, Lu Yuan, Lei Zhang, Jenq-Neng Hwang, et al
[7]

Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Qing Jiang, Chunyuan Li, Jianwei Yang, Hang Su, et al. 2025. Grounding dino: Marry- ing dino with grounded pre-training for open-set object detection. In European Conference on Computer Vision . Springer, 38–55

2025
[8]

OpenAI. 2024. GPT-4 Technical Report. arXiv:2303.08774 [cs.CL] https://arxiv. org/abs/2303.08774

arXiv 2024

Show all 18 references

[9]

Honglin Pang, Xiujin Qi, Chengjun Xiao, Ziying Xu, Guangchen Ding, Yi Chang, Xi Yang, and Tianjing Duan. 2024. Pottery evolution pattern discovery based on deep learning: case study of Miaozigou culture in China. Heritage Science 12, 1 (2024), 352

2024
[10]

Vincent Perot, Kai Kang, Florian Luisier, Guolong Su, Xiaoyu Sun, Ramya Sree Boppana, Zilong Wang, Zifeng Wang, Jiaqi Mu, Hao Zhang, et al. 2023. Lmdx: Language model-based document information extraction and localization. arXiv preprint arXiv:2309.10952 (2023)

2023 arXiv
[11]

Tianhe Ren, Qing Jiang, Shilong Liu, Zhaoyang Zeng, Wenlong Liu, Han Gao, Hongjie Huang, Zhengyu Ma, Xiaoke Jiang, Yihao Chen, et al. 2024. Grounding DINO 1.5: Advance the" Edge" of Open-Set Object Detection. arXiv preprint arXiv:2405.10300 (2024)

2024 arXiv
[12]

Yi Tang, Chia-Ming Chang, and Xi Yang. 2024. PDFChatAnnotator: A Human- LLM Collaborative Multi-Modal Data Annotation Tool for PDF-Format Catalogs. In Proceedings of the 29th International Conference on Intelligent User Interfaces . 419–430

2024
[13]

Haolun Wu, Ye Yuan, Liana Mikaelyan, Alexander Meulemans, Xue Liu, James Hensman, and Bhaskar Mitra. 2024. Structured Entity Extraction Using Large Language Models. arXiv preprint arXiv:2402.04437 (2024)

2024 arXiv
[14]

Shi Yu, Chaoyue Tang, Bokai Xu, Junbo Cui, Junhao Ran, Yukun Yan, Zheng- hao Liu, Shuo Wang, Xu Han, Zhiyuan Liu, et al . 2024. Visrag: Vision-based retrieval-augmented generation on multi-modality documents. arXiv preprint arXiv:2410.10594 (2024)

2024 arXiv
[15]

Haotian Zhang, Pengchuan Zhang, Xiaowei Hu, Yen-Chun Chen, Liunian Li, Xiyang Dai, Lijuan Wang, Lu Yuan, Jenq-Neng Hwang, and Jianfeng Gao. 2022. Glipv2: Unifying localization and vision-language understanding. Advances in Neural Information Processing Systems 35 (2022), 36067–36080

2022
[16]

Rixin Zhou, Jiafu Wei, Qian Zhang, Ruihua Qi, Xi Yang, and Chuntao Li. 2023. Multi-Granularity Archaeological Dating of Chinese Bronze Dings Based on a Knowledge-Guided Relation Graph. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 3103–3113

2023
[2022]

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

Grounded language-image pre-training. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 10965–10975
[2023]

Applied Sciences 13, 22 (2023), 12208

Using augmented small multimodal models to guide large language models for multimodal relation extraction. Applied Sciences 13, 22 (2023), 12208

2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

Anthropic. 2024. Claude 3.5 Sonnet Model Card Addendum. https://www- cdn.anthropic.com/fed9cc193a14b84131812372d8d5857f8f304c52/Model_Card_ Claude_3_Addendum.pdf

2024

[2] [2]

Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. 2023. Qwen-vl: A frontier large vision- language model with versatile abilities. arXiv preprint arXiv:2308.12966 (2023)

arXiv 2023

[3] [3]

Anjanava Biswas and Wrick Talukdar. 2024. Robustness of Structured Data Extraction from In-Plane Rotated Documents Using Multi-Modal Large Language Models (LLM). Journal of Artificial Intelligence Research (2024)

work page 2024

[4] [4]

Xiang Chen, Chenxi Wang, Yida Xue, Ningyu Zhang, Xiaoyan Yang, Qiang Li, Yue Shen, Lei Liang, Jinjie Gu, and Huajun Chen. 2024. Unified hallucination detection for multimodal large language models. arXiv preprint arXiv:2402.03190 (2024)

arXiv 2024

[5] [5]

Wentao He, Hanjie Ma, Shaohua Li, Hui Dong, Haixiang Zhang, and Jie Feng

work page

[6] [6]

Liunian Harold Li, Pengchuan Zhang, Haotian Zhang, Jianwei Yang, Chunyuan Li, Yiwu Zhong, Lijuan Wang, Lu Yuan, Lei Zhang, Jenq-Neng Hwang, et al

[7] [7]

Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Qing Jiang, Chunyuan Li, Jianwei Yang, Hang Su, et al. 2025. Grounding dino: Marry- ing dino with grounded pre-training for open-set object detection. In European Conference on Computer Vision . Springer, 38–55

2025

[8] [8]

OpenAI. 2024. GPT-4 Technical Report. arXiv:2303.08774 [cs.CL] https://arxiv. org/abs/2303.08774

arXiv 2024