RLHF-V: Towards Trustworthy MLLMs via Behavior Alignment from Fine-Grained Correctional Human Feedback
Open accessThe DOI matches its Crossref record; 33 of the 33 references it cites were found in a registry or an index.
Abstract
Multimodal Large Language Models (MLLMs) have recently demonstrated impressive capabilities in multimodal understanding, reasoning, and interaction. However, existing MLLMs prevalently suffer from serious hallucination problems, generating text that is not factually grounded in associated images. The problem makes existing MLLMs untrustworthy and thus impractical in real-world (especially high-stakes) applications. To address the challenge, we present RLHF-V, which enhances MLLM trustworthiness via behavior alignment from fine-grained correctional human feedback. Specifically, RLHF-V collects human preference in the form of segment-level corrections on hallucinations, and performs dense direct preference optimization over the human feedback. Comprehensive experiments on five benchmarks in both automatic and human evaluation show that, RLHF-V can enable substantially more trustworthy MLLM behaviors with promising data and computation efficiency. Remarkably, using 1.4k annotated data samples, RLHF-V significantly reduces the hallucination rate of the base MLLM by 34.8%, outperforming the concurrent LLaVA-RLHF trained on 10k annotated data. The final model achieves state-of-the-art performance in trustwor-thiness among open-source MLLMs, and shows better ro-bustness than GPT-4V in preventing hallucinations aroused from over-generalization.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Microsoft COCO: Common Objects in Context
Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-10602-1_48 · Cited by 45,460
Learning Transferable Visual Models From Natural Language Supervision
Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. (2021) · arXiv · doi:10.48550/arxiv.2103.00020 · Cited by 15,293 · cited 1×
“Data augmentation is widely adopted to improve the data diversity and model robustness in various multi-modal models [11, 19, 32, 41, 47].”
BackgroundProximal Policy Optimization Algorithms
Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017) · arXiv · doi:10.48550/arxiv.1707.06347 · Cited by 10,625 · cited 1×
“It trains the reward model on 10k human-labeled preference data, and performs proximal policy optimization [36] on 72k factually augmented data.”
BackgroundLLaMA: Open and Efficient Foundation Language Models
Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., et al. (2023) · arXiv · doi:10.48550/arxiv.2302.13971 · Cited by 4,602
Training language models to follow instructions with human feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · arXiv · doi:10.48550/arxiv.2203.02155 · Cited by 4,208
Kuznetsova, A., Rom, H., Alldrin, N., Uijlings, J., Krasin, I., Pont-Tuset, J., Kamali, S., Popov, S., Malloci, M., Kolesnikov, A., et al. (2018) · arXiv · doi:10.48550/arxiv.1811.00982 · Cited by 3,653
Training Language Models to Follow Instructions with Human Feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-2011 · Cited by 3,117
Llama 2: Open Foundation and Fine-Tuned Chat Models
Touvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y., Bashlykov, N., Batra, S., Bhargava, P., Bhosale, S., et al. (2023) · arXiv · doi:10.48550/arxiv.2307.09288 · Cited by 2,604
Liu, H., Li, C., Wu, Q., & Lee, Y. J. (2023) · Advances in Neural Information Processing Systems 36 · doi:10.52202/075280-1516 · Cited by 2,428
Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering
Goyal, Y., Khot, T., Summers-Stay, D., Batra, D., & Parikh, D. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.670 · Cited by 2,287
Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena
Zheng, L., Chiang, W.-L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E. P., et al. (2023) · arXiv · doi:10.48550/arxiv.2306.05685 · Cited by 1,888 · cited 1×
“5 [50] (fine-tuned from LLaMA-2 [39]) as LLM backbone.”
Self-citationImproved Baselines with Visual Instruction Tuning
Liu, H., Li, C., Li, Y., & Lee, Y. J. (2023) · arXiv · doi:10.48550/arxiv.2310.03744 · Cited by 1,733
Flamingo: a Visual Language Model for Few-Shot Learning
Alayrac, J.-B., Donahue, J., Luc, P., Miech, A., Barr, I., Hasson, Y., Lenc, K., Mensch, A., Millican, K., Reynolds, M., et al. (2022) · arXiv · doi:10.48550/arxiv.2204.14198 · Cited by 1,555
LAION-5B: An open large-scale dataset for training next generation image-text models
Schuhmann, C., Beaumont, R., Vencu, R., Gordon, C., Wightman, R., Cherti, M., Coombes, T., Katta, A., Mullis, C., Wortsman, M., et al. (2022) · arXiv · doi:10.48550/arxiv.2210.08402 · Cited by 1,022
Li, J., Li, D., Savarese, S., & Hoi, S. (2023) · arXiv · doi:10.48550/arxiv.2301.12597 · Cited by 908 · cited 4×
“The recent success of Multimodal Large Language Models (MLLMs) marks a significant milestone in AI research [2, 4, 11, 12, 19, 27, 29, 42, 51].”
BackgroundInfluentialDirect Preference Optimization: Your Language Model is Secretly a Reward Model
Rafailov, R., Sharma, A., Mitchell, E., Manning, C. D., Ermon, S., & Finn, C. (2023) · Advances in Neural Information Processing Systems 36 · doi:10.52202/075280-2338 · Cited by 686
Liu, H., Li, C., Wu, Q., & Lee, Y. J. (2023) · arXiv · doi:10.48550/arxiv.2304.08485 · Cited by 682
PaLM-E: An Embodied Multimodal Language Model
Driess, D., Xia, F., Sajjadi, M. S. M., Lynch, C., Chowdhery, A., Ichter, B., Wahid, A., Tompson, J., Vuong, Q., Yu, T., et al. (2023) · arXiv · doi:10.48550/arxiv.2303.03378 · Cited by 591 · cited 2×
“The recent success of Multimodal Large Language Models (MLLMs) marks a significant milestone in AI research [2, 4, 11, 12, 19, 27, 29, 42, 51].”
BackgroundSelf-citationImage as a Foreign Language: BEIT Pretraining for Vision and Vision-Language Tasks
Wang, W., Bao, H., Dong, L., Bjorck, J., Peng, Z., Liu, Q., Aggarwal, K., Mohammed, O. K., Singhal, S., Som, S., et al. (2023) · 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52729.2023.01838 · Cited by 561
MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models
Zhu, D., Chen, J., Shen, X., Li, X., & Elhoseiny, M. (2023) · arXiv · doi:10.48550/arxiv.2304.10592 · Cited by 469 · cited 1×
“The recent success of Multimodal Large Language Models (MLLMs) marks a significant milestone in AI research [2, 4, 11, 12, 19, 27, 29, 42, 51].”
BackgroundTraining a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
Bai, Y., Jones, A., Ndousse, K., Askell, A., Chen, A., DasSarma, N., Drain, D., Fort, S., Ganguli, D., Henighan, T., et al. (2022) · arXiv · doi:10.48550/arxiv.2204.05862 · Cited by 442 · cited 2×
“Moreover, mis-allocation of credit to the non-robust bias correlated with the data usually leads to reward hacking and behavior degeneration problems [5, 39].”
BackgroundObject Hallucination in Image Captioning
Rohrbach, A., Hendricks, L. A., Burns, K., Darrell, T., & Saenko, K. (2018) · Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d18-1437 · Cited by 420 · cited 1×
“For trustworthiness, we perform evaluation on three benchmarks: (1) Object HalBench [34] is a widely adopted benchmark for assessing object hallucination in detailed image descriptions.”
MethodDirect Preference Optimization: Your Language Model is Secretly a Reward Model
Rafailov, R., Sharma, A., Mitchell, E., Ermon, S., Manning, C. D., & Finn, C. (2023) · arXiv · doi:10.48550/arxiv.2305.18290 · Cited by 293
Language Is Not All You Need: Aligning Perception with Language Models
Huang, S., Dong, L., Wang, W., Hao, Y., Singhal, S., Ma, S., Lv, T., Cui, L., Mohammed, O. K., Patra, B., et al. (2023) · arXiv · doi:10.48550/arxiv.2302.14045 · Cited by 235 · cited 2×
“Models are pretrained on large-scale image-text pairs [4, 11, 19, 42, 47] or interleaved data [2, 3, 16] to learn the semantic mapping between visual and text signals.”
BackgroundCogVLM: Visual Expert for Pretrained Language Models
Wang, W., Lv, Q., Yu, W., Hong, W., Qi, J., Wang, Y., Ji, J., Yang, Z., Zhao, L., Song, X., et al. (2023) · arXiv · doi:10.48550/arxiv.2311.03079 · Cited by 217
Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond
Bai, J., Bai, S., Yang, S., Wang, S., Tan, S., Wang, P., Lin, J., Zhou, C., & Zhou, J. (2023) · arXiv · doi:10.48550/arxiv.2308.12966 · Cited by 193 · cited 5×
“Models are pretrained on large-scale image-text pairs [4, 11, 19, 42, 47] or interleaved data [2, 3, 16] to learn the semantic mapping between visual and text signals.”
BackgroundMethodInfluential(2023) · doi:10.48550/arxiv.2310.14566 · Cited by 166
Detecting and Preventing Hallucinations in Large Vision Language Models
Gunjal, A., Yin, J., & Bas, E. (2024) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v38i16.29771 · Cited by 161
Mitigating Hallucination in Large Multi-Modal Models via Robust Instruction Tuning
Liu, F., Lin, K., Li, L., Wang, J., Yacoob, Y., & Wang, L. (2023) · arXiv · doi:10.48550/arxiv.2306.14565 · Cited by 149
Scalable agent alignment via reward modeling: a research direction
Leike, J., Krueger, D., Everitt, T., Martic, M., Maini, V., & Legg, S. (2018) · arXiv · doi:10.48550/arxiv.1811.07871 · Cited by 121 · cited 1×
“[15] distinguishes the inaccurate parts in responses via human annotation, and internally discourages the halluci-nated parts by direct preference optimization.”
BackgroundKenton, Z., Everitt, T., Weidinger, L., Gabriel, I., Mikulik, V., & Irving, G. (2021) · arXiv · doi:10.48550/arxiv.2103.14659 · Cited by 45
UltraFeedback: Boosting Language Models with Scaled AI Feedback
Cui, G., Yuan, L., Ding, N., Yao, G., He, B., Zhu, W., Ni, Y., Xie, G., Xie, R., Lin, Y., et al. (2023) · arXiv · doi:10.48550/arxiv.2310.01377 · Cited by 23
Self-citationMathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts
Lu, P., Bansal, H., Xia, T., Liu, J., Li, C., Hajishirzi, H., Cheng, H., Chang, K.-W., Galley, M., & Gao, J. (2023) · arXiv · doi:10.48550/arxiv.2310.02255 · Cited by 12
Cited by
A survey on multimodal large language models
Yin, S., Fu, C., Zhao, S., Li, K., Sun, X., Xu, T., & Chen, E. (2024) · National Science Review · doi:10.1093/nsr/nwae403 · Cited by 893
Zhang, Y., Pan, Y., Zhong, T., Dong, P., Xie, K., Liu, Y., Jiang, H., Wu, Z., Liu, Z., Zhao, W., et al. (2024) · Meta-Radiology · doi:10.1016/j.metrad.2024.100103 · Cited by 52
Self-citationHallucination of Multimodal Large Language Models: A Survey
Bai, Z., Wang, P., Xiao, T., He, T., Han, Z., Zhang, Z., & Shou, M. Z. (2024) · arXiv · doi:10.48550/arxiv.2404.18930 · Cited by 38
“Similarly, RLHF-V [119] also employs the RLHF paradigm to enhance the pre-trained MLLM. Specifically, this work emphasizes two improvements: 1) at the data level, it proposes to collect human feedback in the form of fine-grained segment-level corrections, providing a clear, dense, and fine-grained…”
Self-citationCycle Consistency as Reward: Learning Image-Text Alignment Without Human Preferences
Bahng, H., Chan, C., Durand, F., & Isola, P. (2025) · 2025 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.02129 · Cited by 3
Stop learning it all to mitigate visual hallucination, Focus on the hallucination target
Yoon, D., Song, Y., & Park, W. (2025) · 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52734.2025.00397 · Cited by 1
Metadata
- Authors
- Yu, T., Yao, Y., Zhang, H., He, T., Han, Y., Cui, G., Hu, J., Liu, Z., Zheng, H.-T., & Sun, M.
- Venue
- 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
- Publisher
- IEEE
- Published
- 2024-06-16
- Pages
- 13807-13816
- ISSN
- 1063-6919 (ISSN-L)
- OpenAlex
- W4402727405
- Cited by
- 113
- References
- 33
- FWCI
- 24.67
- Type
- conference paper
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.