Pith. sign in

REVIEW 5 major objections 8 minor 1 cited by

Decompose and Leverage Preferences from Expert Models for Improving Trustworthiness of MLLMs

T0 review · 5 major / 8 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read This paper builds a preference dataset from decomposed, expert-verified claims and shows it cuts MLLM hallucination rates by about fivefold without losing informativeness.

arxiv 2411.13697 v1 pith:LUPRPSU5 submitted 2024-11-20 cs.CV

Rui Cao , Yuming Jiang , Michael Schlichtkrull , Andreas Vlachos This is my paper · ORCID
classification cs.CV
keywords multimodallargelanguagemodelshallucinationpreferenceoptimizationdirectresponsedecompositionexperttrustworthinessopen-vocabularydetection
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

Multimodal large language models (MLLMs) often produce detailed image descriptions that contain confident factual errors. The paper argues that the best way to fix this is to stop asking a single evaluation model to judge a whole response, and instead break every response into small, checkable claims, verify each claim with a specialist open-source model, and use those fine-grained verdicts to build a preference dataset (DGPref) for direct preference optimization. Across two 7B MLLMs, alignment with DGPref reduces object-hallucination rates by roughly an order of magnitude on the response-level metric (LLaVA from 54.7 to 10.3 CHAIRs; Qwen-VL-Chat from 36.0 to 8.0) while keeping or improving informativeness scores. If these results hold, preference-data construction for trustworthy multimodal models no longer requires expensive closed-source judges.

What carries the argument

The load-bearing mechanism is DecompGen's two-stage decomposition. In stage one, an instruction-tuned LLM (Llama-3.1-8B-Instruct) uses eight in-context examples per aspect to split a response into check-worthy parts. In stage two, each part is reduced to atomic tasks via a fixed mapping (e.g., relation verification becomes existence checks for both entities plus a relationship check), and each task is routed to a dedicated expert: OWL-ViT for [DET], BLIP-2 for [RELA]general and [ATTR]general, EasyOCR for [OCR], and hand-coded geometric rules for spatial relations and size-aware attributes. Each expert emits a score of 0 (faithful) or -1 (problematic), the scores are weighted and averaged into an overall response score, and the resulting preferences are used for Direct Preference Optimization.

What would settle it

Take a random sample of, say, 200 MLLM responses from a different image distribution, have annotators label every extracted claim and the preferred response, and then measure DecompGen's per-claim agreement and pairwise ranking accuracy against those labels; if per-claim agreement falls far below the 76-92% the paper reports on its small check, or if the pairwise preference accuracy is near chance, the method's hallucination reductions should fail to replicate on that distribution.

Watch

Extended reading notes

Core claim

The central claim is that a decomposable, interpretable feedback generator can outperform a monolithic judge for preference learning. DecompGen first prompts an LLM to extract 'check-worthy parts' from a response across five aspects—object existence, relations, attributes, counts, and image text—and then maps each part to atomic verification tasks ([DET], [RELA], [ATTR], [COUNT], [OCR]). Each atomic task is executed by a specialist: an open-vocabulary detector (OWL-ViT), a VQA model (BLIP-2) for general relations and attributes, coordinate/size heuristics for spatial relations and counts, and an OCR tool for scene text. The binary pass/fail scores for every part are aggregated into a single response score, and pairwise comparisons form DGPref, consisting of 52k preference pairs from 16k Visual Genome images. After DPO with LoRA, LLaVA-v1.5 and Qwen-VL-Chat show large drops in hallucination on ObjHal (CHAIRs 10.3 and 8.0) and lower or comparable error rates than preference-data baselines built with GPT-4 or a 34B evaluator, while preserving coverage and informativeness on MMHal and AMBER.

Load-bearing premise

The pipeline's value depends on the LLM's extraction and the expert models' verdicts being reliable enough that the binary scores separate faithful from hallucinated content, and the paper's own manual checks show this foundation is imperfect (OCR 50%, invalid extraction 77% of decomposition errors).

Editorial extensions

If this is right

  • Preference data for MLLM alignment can be generated entirely from open-source components, eliminating the need for closed-source judges like GPT-4 in this pipeline.
  • Each error in a model response becomes traceable to a specific extracted claim and the expert that verified it, making feedback interpretable and easy to audit.
  • The modular expert design means stronger or additional specialists (e.g., for commonsense or fine-grained OCR) can be swapped in to improve the preference data without redesigning the framework.
  • Because automatic generation scales, DGPref (52k pairs) can exceed the size of human-annotated preference sets (~1k-10k pairs), which the paper shows correlates with better alignment outcomes.

Reading between the lines

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

  • The same decompose-and-verify recipe could be applied to other long-form generation tasks (e.g., text-only summarization or statement verification): any output that can be split into atomic claims and checked by a portfolio of tools could yield preference data without training a monolithic judge. This is an extension the paper does not make.
  • The paper's own error analysis (OCR expert at 50% accuracy; invalid extraction as 77% of decomposition errors) suggests that the main headroom for further gains lies in better extraction and text verification; if those components improve, DGPref's quality and the resulting hallucination reductions should improve correspondingly—a testable prediction.
  • The fact that aligning with ground-truth annotations (PrefObj-GT) improved ObjHal but hurt generalization to MMHal/AMBER cautions that expert-based labels, while noisy, can be more diverse and robust than sparse human annotations; this is a nuance the paper surfaces but does not fully resolve.
  • The reported gains are measured on three specific hallucination benchmarks; treating 'trustworthiness' as a broader property would require testing on additional dimensions such as safety, bias, and out-of-distribution images.
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.

Referee Report

5 major / 8 minor

Summary. The paper proposes DecompGen, a decomposable framework that evaluates MLLM image-description responses by first decomposing each response into visually check-worthy parts (objects, relations, attributes, counts, image text) and then verifying each part through a set of open-source expert models (OWL-ViT, BLIP-2, EasyOCR, heuristic rules). The resulting fine-grained feedback scores are aggregated into an overall score, which is used to construct a preference dataset, DGPref, by pairing higher- and lower-scoring responses. MLLMs (LLaVA-v1.5 and Qwen-VL-Chat at 7B) are then aligned with DGPref via DPO with LoRA. The main empirical claim is that this alignment sharply reduces hallucination rates on ObjHal, MMHal, and AMBER while preserving informativeness, and that DGPref-aligned models outperform or match prior preference-data methods such as Silkie, HSA-DPO, and RLAIF-V.

Significance. If the central claim holds, the paper makes a useful contribution: it replaces closed-source evaluator feedback (e.g., GPT-4) with an ensemble of open-source experts, at a total parameter count of roughly 4B, and it shows large hallucination reductions on two different base MLLMs. The decomposition into atomic verification tasks is interpretable, and the framework allows swapping individual expert models. The paper also releases code, reports a human evaluation, and provides an ablation study with an interesting negative result (PrefObj-GT fails to generalize to MMHal/AMBER). The significance is conditional, however, on the reliability of the automatically generated preference labels and on the independence of the evaluation benchmarks from the training image distribution; both of these points need to be established before the reported gains can be fully credited.

major comments (5)
  1. [Sec. 2.3, Sec. 3.4, Appendix E Table 6] The load-bearing premise—that DecompGen's per-part scores correctly separate faithful from hallucinated response content—is not directly validated. Appendix E Table 6 reports expert accuracies on only 50 manually checked cases, with OCR at 50.0%, general attribute at 76.2%, and general relation at 79.4%. Section 3.4 adds that invalid check-worthy part extraction is the most common decomposition error, accounting for 77% of errors. Because each DGPref preferred/rejected label is a deterministic function of these scores, label noise is inherited from both stages. The paper, however, reports no evaluation of the resulting pairwise preference labels—e.g., no human agreement on which response in a DGPref pair is more accurate, and no analysis of how many pairs would flip if the noisy parts are removed or corrected. I request (a) a human validation of a random sample of DGPref pairs, (b) a per-error-propagation analysis showing the fraction of pairs whose labels change when OCR/general-attribute/general-relation errors are corrected, and (c) an experiment training on a variant of DGPref that excludes or down-weights the least reliable expert dimensions (particularly OCR).
  2. [Tables 2, 3, 5] No error bars, confidence intervals, or significance tests are reported for any evaluation metric. The key comparisons—DGPrefQwen versus HSA-DPO and RLAIF-V, and DGPref versus its ablations—involve differences of a few points or less on several metrics (e.g., MMHal HalRate 0.34 vs. 0.29 for RLAIF-V, AMBER s.CHAIRi 1.5 vs. 3.1 for RLAIF-V). Given the known variance of hallucination benchmarks and the fact that evaluation of MMHal uses GPT-4 with stochastic sampling, these differences may not be robust. Please report results across at least three seeds or bootstrap intervals for the main models and ablations, and state the number of evaluation samples underlying each metric.
  3. [Sec. 3.1, Table 2] The training and evaluation image sources may overlap, which could inflate the reported gains. DGPref is constructed from Visual Genome images, while ObjHal is a COCO-based benchmark and AMBER may also draw on common image sources; Visual Genome contains many images that also appear in COCO. The paper does not quantify this overlap. If a substantial fraction of evaluation images appear in the DGPref training set, the CHAIRs/CHAIRi reductions on ObjHal (e.g., from 36.0 to 8.0 for Qwen) would partly reflect in-distribution leakage rather than improved trustworthiness. Please report the exact image overlap between DGPref and each evaluation benchmark, and provide results on a strictly disjoint subset of each benchmark.
  4. [Sec. 2.3] The aggregation step that converts per-part scores into an overall response score is not specified. The text says 'Weighted averaging is applied over the fine-grained feedback scores' but does not give the formula, the weights, how skipped verifications (where existence fails) are treated, or how the number of check-worthy parts per response is normalized. This is not a cosmetic omission: the preferred/rejected ordering of every DGPref pair depends on this aggregation, and without it the dataset construction is not reproducible or auditable. Please provide the exact aggregation function, the weight values, and the treatment of skipped parts.
  5. [Appendix M, Table 4] The human evaluation that supports the claim that trustworthiness improvements do not sacrifice informativeness is based on very low inter-annotator reliability: Cohen's kappa is 0.13 for 'fewer errors', 0.15 for 'more informative', and 0.24 for 'preferable'. With these levels of agreement, the reported 81.7–88.3% preference rates are difficult to interpret. The human evaluation should either use more reliable annotation procedures (e.g., adjudication, more annotators, clearer criteria) or be reported with appropriate caveats, and it should not serve as the primary evidence against the 'conservatism/length artifact' concern without a length-controlled analysis.
minor comments (8)
  1. [Abstract / Sec. 1] There is a duplicated word: 'image content content' in the Introduction.
  2. [Sec. 3.2] Typo: 'Fianally' should be 'Finally'.
  3. [Sec. 3.2] Typo: 'HAS-DPO' should be 'HSA-DPO' in the sentence comparing with the strongest baselines.
  4. [Sec. 3.2 / Table 2 caption] Typo: 'comparbale' should be 'comparable'.
  5. [Fig. 3] The figure caption contains 'Hallicination' which should be 'Hallucination'.
  6. [Sec. 2.3] The dataset name 'DGPref' is said to stand for 'DecompGen Feedback Generated Preferance dataset'—'Preferance' is misspelled; it should be 'Preference'.
  7. [Appendix A] The reference to BLIP-2 in the description of [RELA]general cites [46] (Woodpecker) in one place and [20] (BLIP-2) elsewhere; the reference numbers appear inconsistent. Please verify all expert-model citations.
  8. [Sec. 3.1 / Appendix I] Please clarify whether the 52k DGPref samples are all distinct pairs per image and how the '8 diverse instructions' are paired with images; the paper says 'pairwise combinations' but the numbers (16k images, 8 responses) imply 448k possible pairs, so the sampling procedure should be stated explicitly.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: DGPref preference labels come from external expert feedback, evaluation uses independent benchmarks, and the only author-overlapping citation is a non-load-bearing methodological reference.

full rationale

The claimed derivation chain is empirical rather than formal: DecompGen produces per-part scores from external expert tools (OWL-ViT for detection, EasyOCR for OCR, BLIP-2 for VQA-based attribute and relation checks, plus explicit heuristic rules), DGPref labels are weighted averages of those scores, DPO training consumes those labels, and trustworthiness is measured on ObjHal, MMHal, and AMBER. No equation in the paper equates the training signal to the evaluation metric, and no parameter is fitted to the benchmark and then reported as a prediction. The preference dataset is built from Visual Genome images, not from the benchmark images or annotations, so the main improvements are not forced by construction. The only author-overlapping citation is [5] (Cao and Jiang, ACL Findings 2023), used for the general idea of decomposition and for separating spatial-relation heuristics; the paper supplies its own prompt templates, Table 1 mapping, and Algorithms 1-2, so this self-citation is not load-bearing. The manuscript itself flags validity limitations: Appendix E Table 6 reports expert accuracies including 50.0% for OCR and 79.4% for general relations on a 50-case check, and Section 3.4 reports that invalid check-worthy part extraction is 77% of decomposition errors. These are correctness and robustness concerns about feedback quality, not evidence that the results reduce to their inputs. One localized overlap is the PrefObj-GT ablation, which uses COCO ground-truth object annotations for feedback and is then evaluated on ObjHal, a benchmark that also uses COCO GT object annotations; the paper explicitly acknowledges this ('which also leverages GT object annotations for evaluation') and does not use that ablation as the central claim, instead showing that it fails to generalize to MMHal and AMBER. Overall, the central claim that DGPref improves trustworthiness is independently testable and is not circular; the score of 2 reflects only the minor, non-load-bearing self-citation.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or theoretical entities. The central claim rests on design assumptions about which content is check-worthy, whether expert models are accurate enough, whether LLM decomposition is reliable, and whether DPO on the resulting pairs improves trustworthiness. These are domain assumptions rather than free parameters in a derivation, though several thresholds and hyperparameters are hand-set and influence the result.

free parameters (6)
  • OWL-ViT detection confidence threshold = 0.25 (ablated to 0.1)
    Controls which objects count as detected; Section 3.3 shows lowering it to 0.1 degrades preference data quality. Chosen empirically.
  • Grammar filter threshold = 0.75
    Filters generated yes/no questions by grammar score before expert verification (Appendix I); a hand-set cutoff that affects feedback coverage.
  • Size attribute heuristic thresholds = 0.4/0.3/0.5 of image height or width
    Algorithm 2 defines large/small/long/short/tall via box-size thresholds; these rules directly determine attribute feedback scores.
  • 'Near' spatial relation threshold = 10% of image width or height
    Algorithm 1 declares 'near' when center coordinate difference is below 0.1 times image width or height; a hand-set constant.
  • Feedback aggregation weights = not reported
    Section 2.3 says weighted averaging is applied to fine-grained scores, but the weights are not given; they determine preference ordering and therefore affect training.
  • DPO and LoRA hyperparameters = beta=0.1, LoRA rank 32, alpha 32, lr 5e-6 or 7e-6, epochs 2 or 1
    Appendix I; training choices affect aligned model quality but are not fitted to the evaluation benchmarks.
assumptions (4)
  • domain assumption A response's trustworthiness can be assessed by checking only five visually check-worthy aspects: object existence, object relations, object attributes, counts of objects, and image text.
    Adopted from prior work [7,50] in Section 2.1; if hallucinations live outside these aspects, DGPref feedback will not capture them.
  • domain assumption Expert model outputs are accurate enough proxies for ground truth in preference labeling.
    OWL-ViT, EasyOCR, BLIP-2, and heuristics are used without per-sample verification; Appendix E estimates OCR accuracy at 50.0% and general attribute/relation accuracy at 76.2%/79.4% on 50 manually checked cases.
  • domain assumption The LLM decomposition (using Llama-3.1-8B) extracts check-worthy parts with an acceptable error rate.
    Section 3.4 reports a low miss rate (under 10%) but invalid extractions are the most common error type, 77% of errors; errors propagate to feedback scores and preference labels.
  • domain assumption Preference pairs derived from aggregate feedback scores are valid training signals for DPO.
    Pairs are formed from weighted averages of fine-grained scores (Section 2.3); there is no verification that aggregate ordering matches human preferences, and the human evaluation in Appendix M shows low inter-annotator agreement.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Decompose and Leverage Preferences from Expert Models for Improving Trustworthiness of MLLMs." pith.science (2026). https://pith.science/paper/LUPRPSU5

@misc{pith2026241113697,
  author       = {Pith},
  title        = {Pith review of: Decompose and Leverage Preferences from Expert Models for Improving Trustworthiness of MLLMs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LUPRPSU5}},
  note         = {Machine review of arXiv:2411.13697}
}
read the original abstract

Multimodal Large Language Models (MLLMs) can enhance trustworthiness by aligning with human preferences. As human preference labeling is laborious, recent works employ evaluation models for assessing MLLMs' responses, using the model-based assessments to automate preference dataset construction. This approach, however, faces challenges with MLLMs' lengthy and compositional responses, which often require diverse reasoning skills that a single evaluation model may not fully possess. Additionally, most existing methods rely on closed-source models as evaluators. To address limitations, we propose DecompGen, a decomposable framework that uses an ensemble of open-sourced expert models. DecompGen breaks down each response into atomic verification tasks, assigning each task to an appropriate expert model to generate fine-grained assessments. The DecompGen feedback is used to automatically construct our preference dataset, DGPref. MLLMs aligned with DGPref via preference learning show improvements in trustworthiness, demonstrating the effectiveness of DecompGen.

Figures

Figures reproduced from arXiv: 2411.13697 by the authors.

Figure 1
Figure 1. A problematic response about detailed image de￾scriptions from LLaVA-v1.5. We proposed DecompGen, a de￾composable and interpretable framework to assess MLLMs’ re￾sponses. It decomposes responses to atomic verification tasks, assigns atomic tasks to proper expert models and generate fine￾grained feedback with an ensemble of expert models. Expert K stands for the k-th expert model. of responses anchoring the visual co… view at source ↗
Figure 2
Figure 2. Overview of the DecompGen model for feedback generation. It extracts visually check-worthy parts of a model response, decomposes each part into atomic verification tasks and assembles and executes expert models for fine-grained feedback generation. C-W Asp. Atomic Tasks Obj. Existence [DET](obj) Obj. Relation [RELA]([DET](objs),rela,[DET](objo)) Obj. Attribute [ATTR](attr, [DET](obj)) Counts of Obj. [COUNT](num, [DE… view at source ↗
Figure 3
Figure 3. Performance of Qwen-VL-Chat on ObjHal after preference learning with different amount of data. Hallicina￾tion rates drops with the increase of data amount. row of the second block in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Visualization of responses and generated feedback. Skipped verifications are not shown in the figure. Model Fewer Errors More Inform. Prefer Base. 18.3% 16.7% 11.7% +DGPref 81.7% 83.3% 88.3% [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Annotation instructions provided to the annotators and [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Eight diverse instructions for generating detailed image [PITH_FULL_IMAGE:figures/full_fig_p021_6.png]
Figure 7
Figure 7. Figure 7: The prompt template for extracting object entities in a response. [PITH_FULL_IMAGE:figures/full_fig_p022_7.png]
Figure 8
Figure 8. Figure 8: The prompt template for extracting object relations in a response. [PITH_FULL_IMAGE:figures/full_fig_p023_8.png]
Figure 9
Figure 9. Figure 9: The prompt template for extracting object attributes in a response. [PITH_FULL_IMAGE:figures/full_fig_p024_9.png]
Figure 10
Figure 10. Figure 10: The prompt template for extracting the counts of an object in a response. [PITH_FULL_IMAGE:figures/full_fig_p025_10.png]
Figure 11
Figure 11. Figure 11: The prompt template for extracting mentioned image texts in a response. [PITH_FULL_IMAGE:figures/full_fig_p026_11.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. InternLM-XComposer2.5-Reward: A Simple Yet Effective Multi-Modal Reward Model

    cs.CV 2025-01 conditional novelty 6.0 of 10

    IXC-2.5-Reward is an open-source multimodal reward model that achieves 70.0% macro accuracy on VL-RewardBench and improves LVLM chat via PPO.

Reference graph

Works this paper leans on

54 extracted references · 35 canonical work pages · cited by 1 Pith paper

  1. [1]

    Llama 3.1 8b instruct

    Meta AI. Llama 3.1 8b instruct. https : / / huggingface.co/meta-llama/Llama-3.1-8B- Instruct, 2024. Accessed: 2024-11-14. 5

    work page 2024

  2. [2]

    Neural module networks

    Jacob Andreas, Marcus Rohrbach, Trevor Darrell, and Dan Klein. Neural module networks. In 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR , pages 39–48, 2016. 2

    work page 2016

  3. [3]

    Qwen-vl: A frontier large vision-language model with versatile abilities

    Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. Qwen-vl: A frontier large vision-language model with versatile abilities. CoRR, abs/2308.12966, 2023. 4, 5, 14

    arXiv 2023

  4. [4]

    Rank analysis of incomplete block designs: I

    Ralph Allan Bradley and Milton E Terry. Rank analysis of incomplete block designs: I. the method of paired compar- isons. Biometrika, 39(3/4):324–345, 1952. 4

    work page 1952

  5. [5]

    Modularized zero-shot VQA with pre-trained models

    Rui Cao and Jing Jiang. Modularized zero-shot VQA with pre-trained models. In Findings of the Association for Com- putational Linguistics: ACL, pages 58–76, 2023. 2, 8, 12

    work page 2023

  6. [6]

    Complex claim verification with evidence re- trieved in the wild

    Jifan Chen, Grace Kim, Aniruddh Sriram, Greg Durrett, and Eunsol Choi. Complex claim verification with evidence re- trieved in the wild. InProceedings of the 2024 Conference of the North American Chapter of the Association for Computa- tional Linguistics: Human Language Technologies (Volume 1: Long Papers), NAACL, pages 3569–3587, 2024. 8

    work page 2024

  7. [7]

    MME: A comprehensive evaluation benchmark for multimodal large language models

    Chaoyou Fu, Peixian Chen, Yunhang Shen, Yulei Qin, Mengdan Zhang, Xu Lin, Zhenyu Qiu, Wei Lin, Jinrui Yang, Xiawu Zheng, Ke Li, Xing Sun, and Rongrong Ji. MME: A comprehensive evaluation benchmark for multimodal large language models. CoRR, abs/2306.13394, 2023. 2

    arXiv 2023

  8. [8]

    Making the V in VQA matter: El- evating the role of image understanding in visual question answering

    Yash Goyal, Tejas Khot, Douglas Summers-Stay, Dhruv Ba- tra, and Devi Parikh. Making the V in VQA matter: El- evating the role of image understanding in visual question answering. In 2017 IEEE Conference on Computer Vision and Pattern Recognition, CVPR, pages 6325–6334, 2017. 1, 16

    work page 2017

Show all 54 references
  1. [9]

    Breaking common sense: Whoops! A vision-and- language benchmark of synthetic and compositional images

    Nitzan Bitton Guetta, Yonatan Bitton, Jack Hessel, Lud- wig Schmidt, Yuval Elovici, Gabriel Stanovsky, and Roy Schwartz. Breaking common sense: Whoops! A vision-and- language benchmark of synthetic and compositional images. In IEEE/CVF International Conference on Computer Visi...

    2023

  2. [10]

    Efficient mul- timodal learning from data-centric perspective

    Muyang He, Yexin Liu, Boya Wu, Jianhao Yuan, Yueze Wang, Tiejun Huang, and Bo Zhao. Efficient mul- timodal learning from data-centric perspective. CoRR, abs/2402.11530, 2024. 1

    2024 arXiv

  3. [11]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. In The Tenth International Conference on Learning Represen- tations, ICLR, 2022. 5

    2022

  4. [12]

    Learning to reason: End-to-end module networks for visual question answering

    Ronghang Hu, Jacob Andreas, Marcus Rohrbach, Trevor Darrell, and Kate Saenko. Learning to reason: End-to-end module networks for visual question answering. In IEEE International Conference on Computer Vision, ICCV, pages 804–813. IEEE Computer Society, 2017. 2

    2017

  5. [13]

    Opera: Alleviating hallucination in multi- modal large language models via over-trust penalty and retrospection-allocation

    Qidong Huang, Xiaoyi Dong, Pan Zhang, Bin Wang, Con- ghui He, Jiaqi Wang, Dahua Lin, Weiming Zhang, and Nenghai Yu. Opera: Alleviating hallucination in multi- modal large language models via over-trust penalty and retrospection-allocation. In IEEE Conference on Computer Vision...

  6. [14]

    Hudson and Christopher D

    Drew A. Hudson and Christopher D. Manning. GQA: A new dataset for real-world visual reasoning and composi- tional question answering. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR , pages 6700–6709,

  7. [15]

    Hallucination augmented contrastive learning for multimodal large language model

    Chaoya Jiang, Haiyang Xu, Mengfan Dong, Jiaxing Chen, Wei Ye, Ming Yan, Qinghao Ye, Ji Zhang, Fei Huang, and Shikun Zhang. Hallucination augmented contrastive learning for multimodal large language model. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR , pa...

    2024

  8. [16]

    Hal-eval: A uni- versal and fine-grained hallucination evaluation framework for large vision language models

    Chaoya Jiang, Wei Ye, Mengfan Dong, Hongrui Jia, Haiyang Xu, Ming Yan, Ji Zhang, and Shikun Zhang. Hal-eval: A uni- versal and fine-grained hallucination evaluation framework for large vision language models. CoRR, abs/2402.15721,

    arXiv

  9. [17]

    What’s ”up” with vision-language models? investigating their strug- gle with spatial reasoning

    Amita Kamath, Jack Hessel, and Kai-Wei Chang. What’s ”up” with vision-language models? investigating their strug- gle with spatial reasoning. In Proceedings of the 2023 Con- ference on Empirical Methods in Natural Language Process- ing, EMNLP, pages 9161–9175, 2023. 2, 12

    2023

  10. [18]

    Shamma, Michael S

    Ranjay Krishna, Yuke Zhu, Oliver Groth, Justin Johnson, Kenji Hata, Joshua Kravitz, Stephanie Chen, Yannis Kalan- tidis, Li-Jia Li, David A. Shamma, Michael S. Bernstein, and Li Fei-Fei. Visual genome: Connecting language and vision using crowdsourced dense image annotations. ...

    2017

  11. [19]

    Mitigating object hallucinations in large vision-language models through vi- sual contrastive decoding

    Sicong Leng, Hang Zhang, Guanzheng Chen, Xin Li, Shi- jian Lu, Chunyan Miao, and Lidong Bing. Mitigating object hallucinations in large vision-language models through vi- sual contrastive decoding. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR, pages 13872–13882,

  12. [20]

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven C. H. Hoi. BLIP-2: bootstrapping language-image pre-training with frozen image encoders and large language models. In International Conference on Machine Learning, ICML , pages 19730–19742, 2023. 1, 3, 8, 12

    2023

  13. [21]

    Silkie: Preference distillation for large visual lan- guage models

    Lei Li, Zhihui Xie, Mukai Li, Shunian Chen, Peiyi Wang, Liang Chen, Yazheng Yang, Benyou Wang, and Lingpeng Kong. Silkie: Preference distillation for large visual lan- guage models. CoRR, abs/2312.10665, 2023. 1, 2, 4, 5, 8, 16

    2023 arXiv

  14. [22]

    Evaluating object hallucination in large vision-language models

    Yifan Li, Yifan Du, Kun Zhou, Jinpeng Wang, Wayne Xin Zhao, and Ji-Rong Wen. Evaluating object hallucination in large vision-language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Pro- cessing, EMNLP, pages 292–305, 2023. 4, 8, 16

    2023

  15. [23]

    Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll ´ar, and C

    Tsung-Yi Lin, Michael Maire, Serge J. Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll ´ar, and C. Lawrence Zitnick. Microsoft COCO: common objects in context. In Computer Vision - ECCV, pages 740–755, 2014. 7

    2014

  16. [24]

    Improved baselines with visual instruction tuning

    Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. Improved baselines with visual instruction tuning. CoRR, abs/2310.03744, 2023. 1, 2, 4, 5, 8, 14, 15, 16

    2023 arXiv

  17. [25]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. InAdvances in Neural Information Processing Systems 36: Annual Conference on Neural Infor- mation Processing Systems NeurIPS, 2023. 1, 8

    2023

  18. [26]

    OK-VQA: A visual question answering benchmark requiring external knowledge

    Kenneth Marino, Mohammad Rastegari, Ali Farhadi, and Roozbeh Mottaghi. OK-VQA: A visual question answering benchmark requiring external knowledge. In IEEE Confer- ence on Computer Vision and Pattern Recognition, CVPR , pages 3195–3204, 2019. 1

    2019

  19. [27]

    Factscore: Fine-grained atomic evaluation of factual precision in long form text generation

    Sewon Min, Kalpesh Krishna, Xinxi Lyu, Mike Lewis, Wen- tau Yih, Pang Wei Koh, Mohit Iyyer, Luke Zettlemoyer, and Hannaneh Hajishirzi. Factscore: Fine-grained atomic evaluation of factual precision in long form text generation. In Proceedings of the 2023 Conference on Empirica...

    2023

  20. [28]

    Matthias Minderer, Alexey A. Gritsenko, Austin Stone, Maxim Neumann, Dirk Weissenborn, Alexey Dosovitskiy, Aravindh Mahendran, Anurag Arnab, Mostafa Dehghani, Zhuoran Shen, Xiao Wang, Xiaohua Zhai, Thomas Kipf, and Neil Houlsby. Simple open-vocabulary object detection with vis...

    2022 arXiv

  21. [29]

    Compositional chain-of-thought prompting for large multimodal models

    Chancharik Mitra, Brandon Huang, Trevor Darrell, and Roei Herzig. Compositional chain-of-thought prompting for large multimodal models. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR, pages 14420–14431,

  22. [30]

    Morris, Eli Lifland, Jin Yong Yoo, Jake Grigsby, Di Jin, and Yanjun Qi

    John X. Morris, Eli Lifland, Jin Yong Yoo, Jake Grigsby, Di Jin, and Yanjun Qi. Textattack: A framework for adver- sarial attacks, data augmentation, and adversarial training in NLP. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: Sys...

    2020

  23. [31]

    GPT-4 technical report

    OpenAI. GPT-4 technical report. CoRR, abs/2303.08774,

    arXiv

  24. [32]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Car- roll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F. Christiano, Jan Le...

    2022

  25. [33]

    Manning, Stefano Ermon, and Chelsea Finn

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christo- pher D. Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. In Advances in Neural Information Process- ing Systems 36: Annual Conference on Neural Info...

    2023

  26. [34]

    Object hallucination in image cap- tioning

    Anna Rohrbach, Lisa Anne Hendricks, Kaylee Burns, Trevor Darrell, and Kate Saenko. Object hallucination in image cap- tioning. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 4035–4045,

    2018

  27. [35]

    Arik, and Tomas Pfister

    Pritam Sarkar, Sayna Ebrahimi, Ali Etemad, Ahmad Beirami, Sercan ¨O. Arik, and Tomas Pfister. Mitigating object hallucination via data augmented contrastive tuning. CoRR, abs/2405.18654, 2024. 5, 7, 8

    2024 arXiv

  28. [36]

    Toolformer: Lan- guage models can teach themselves to use tools

    Timo Schick, Jane Dwivedi-Yu, Roberto Dess `ı, Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. Toolformer: Lan- guage models can teach themselves to use tools. In Ad- vances in Neural Information Processing Systems 36: An- nu...

    2023

  29. [37]

    Averitec: A dataset for real-world claim verification with ev- idence from the web

    Michael Schlichtkrull, Zhijiang Guo, and Andreas Vlachos. Averitec: A dataset for real-world claim verification with ev- idence from the web. In Advances in Neural Information Processing Systems 36: Annual Conference on Neural Infor- mation Processing Systems 2023, NeurIPS, 2023. 8

    2023

  30. [38]

    Towards VQA models that can read

    Amanpreet Singh, Vivek Natarajan, Meet Shah, Yu Jiang, Xinlei Chen, Dhruv Batra, Devi Parikh, and Marcus Rohrbach. Towards VQA models that can read. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR, pages 8317–8326, 2019. 15

    2019

  31. [39]

    Reclip: A strong zero-shot baseline for referring expression compre- hension

    Sanjay Subramanian, William Merrill, Trevor Darrell, Matt Gardner, Sameer Singh, and Anna Rohrbach. Reclip: A strong zero-shot baseline for referring expression compre- hension. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume ...

    2022

  32. [40]

    Aligning large multimodal models with factually aug- mented RLHF

    Zhiqing Sun, Sheng Shen, Shengcao Cao, Haotian Liu, Chunyuan Li, Yikang Shen, Chuang Gan, Liangyan Gui, Yu- Xiong Wang, Yiming Yang, Kurt Keutzer, and Trevor Dar- rell. Aligning large multimodal models with factually aug- mented RLHF. In Findings of the Association for Computa...

    2024

  33. [41]

    Winoground: Probing vision and language models for visio- linguistic compositionality

    Tristan Thrush, Ryan Jiang, Max Bartolo, Amanpreet Singh, Adina Williams, Douwe Kiela, and Candace Ross. Winoground: Probing vision and language models for visio- linguistic compositionality. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR , pages 5228–...

    2022

  34. [42]

    An llm-free multi-dimensional benchmark for mllms hallu- cination evaluation

    Junyang Wang, Yuhang Wang, Guohai Xu, Jing Zhang, Yukai Gu, Haitao Jia, Ming Yan, Ji Zhang, and Jitao Sang. An llm-free multi-dimensional benchmark for mllms hallu- cination evaluation. CoRR, abs/2311.07397, 2023. 1, 4, 8

    2023 arXiv

  35. [43]

    Visual chatgpt: Talking, drawing and editing with visual foundation models

    Chenfei Wu, Shengming Yin, Weizhen Qi, Xiaodong Wang, Zecheng Tang, and Nan Duan. Visual chatgpt: Talking, drawing and editing with visual foundation models. CoRR, abs/2303.04671, 2023. 8

    2023 arXiv

  36. [44]

    Detecting and mitigating hallucination in large vi- sion language models via fine-grained AI feedback

    Wenyi Xiao, Ziwei Huang, Leilei Gan, Wanggui He, Haoyuan Li, Zhelun Yu, Hao Jiang, Fei Wu, and Linchao Zhu. Detecting and mitigating hallucination in large vi- sion language models via fine-grained AI feedback. CoRR, abs/2404.14233, 2024. 1, 2, 5, 7, 8, 16, 21

    2024 arXiv

  37. [45]

    mplug-owl: Modularization empowers large language models with mul- timodality

    Qinghao Ye, Haiyang Xu, Guohai Xu, Jiabo Ye, Ming Yan, Yiyang Zhou, Junyang Wang, Anwen Hu, Pengcheng Shi, Yaya Shi, Chenliang Li, Yuanhong Xu, Hehong Chen, Jun- feng Tian, Qian Qi, Ji Zhang, and Fei Huang. mplug-owl: Modularization empowers large language models with mul- tim...

    2023 arXiv

  38. [46]

    Woodpecker: Hallucination correction for multimodal large language models

    Shukang Yin, Chaoyou Fu, Sirui Zhao, Tong Xu, Hao Wang, Dianbo Sui, Yunhang Shen, Ke Li, Xing Sun, and Enhong Chen. Woodpecker: Hallucination correction for multimodal large language models. CoRR, abs/2310.16045, 2023. 12

    2023 arXiv

  39. [47]

    Rlhf-v: Towards trustworthy mllms via behavior alignment from fine-grained correctional human feedback

    Tianyu Yu, Yuan Yao, Haoye Zhang, Taiwen He, Yifeng Han, Ganqu Cui, Jinyi Hu, Zhiyuan Liu, Hai-Tao Zheng, Maosong Sun, et al. Rlhf-v: Towards trustworthy mllms via behavior alignment from fine-grained correctional human feedback. In IEEE Conference on Computer Vision and Pat- ...

    2024

  40. [48]

    RLAIF-V: aligning mllms through open-source AI feedback for super GPT-4V trustworthiness

    Tianyu Yu, Haoye Zhang, Yuan Yao, Yunkai Dang, Da Chen, Xiaoman Lu, Ganqu Cui, Taiwen He, Zhiyuan Liu, Tat-Seng Chua, and Maosong Sun. RLAIF-V: aligning mllms through open-source AI feedback for super GPT-4V trustworthiness. CoRR, abs/2405.17220, 2024. 1, 4, 5, 8, 16, 21

    2024

  41. [49]

    Less is more: Miti- gating multimodal hallucination from an EOS decision per- spective

    Zihao Yue, Liang Zhang, and Qin Jin. Less is more: Miti- gating multimodal hallucination from an EOS decision per- spective. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL, pages 11766–11781, 2024. 5, 8

    2024

  42. [50]

    Vl- checklist: Evaluating pre-trained vision-language mod- els with objects, attributes and relations

    Tiancheng Zhao, Tianqi Zhang, Mingwei Zhu, Haozhan Shen, Kyusong Lee, Xiaopeng Lu, and Jianwei Yin. Vl- checklist: Evaluating pre-trained vision-language mod- els with objects, attributes and relations. CoRR, abs/2207.00221, 2022. 2, 12

    2022 arXiv

  43. [51]

    Beyond hallucinations: Enhanc- ing lvlms through hallucination-aware direct preference op- timization

    Zhiyuan Zhao, Bin Wang, Linke Ouyang, Xiaoyi Dong, Ji- aqi Wang, and Conghui He. Beyond hallucinations: Enhanc- ing lvlms through hallucination-aware direct preference op- timization. CoRR, abs/2311.16839, 2023. 1, 2, 4, 5, 8

    2023 arXiv

  44. [52]

    ROME: evaluating pre-trained vision-language models on reasoning beyond visual common sense

    Kankan Zhou, Eason Lai, Wei Bin Au Yeong, Kyriakos Mouratidis, and Jing Jiang. ROME: evaluating pre-trained vision-language models on reasoning beyond visual common sense. In Findings of the Association for Computational Lin- guistics: EMNLP, pages 10185–10197, 2023. 12

    2023

  45. [53]

    Aligning modalities in vision large language models via preference fine-tuning

    Yiyang Zhou, Chenhang Cui, Rafael Rafailov, Chelsea Finn, and Huaxiu Yao. Aligning modalities in vision large language models via preference fine-tuning. CoRR, abs/2402.11411, 2024. 1, 5, 8

    2024 arXiv

  46. [54]

    Is the {sub} {relation} {obj}?

    Yiyang Zhou, Chenhang Cui, Jaehong Yoon, Linjun Zhang, Zhun Deng, Chelsea Finn, Mohit Bansal, and Huaxiu Yao. Analyzing and mitigating object hallucination in large vision-language models. In The Twelfth International Con- ference on Learning Representations, ICLR, 2024. 5, 8 ...

    2024

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.