Elamur Research⌕
☰
home / papers / 2023 / Woodpecker: Hallucination Correction for Multimodal Large Language Models

Woodpecker: Hallucination Correction for Multimodal Large Language Models

Paper · 2023 · Shukang Yin, Chaoyou Fu, Sirui Zhao et al.

Summary

Multimodal large language models (MLLMs) often describe things that are not in the image. They may invent objects or get attributes such as count or color wrong. Earlier fixes like LRV-Instruction and VIGC rely on instruction tuning, which needs new data and expensive retraining, and can make outputs shorter or slower to generate. Woodpecker takes a different route. It is a training-free, post-hoc corrector that edits an MLLM's answer after it has been produced. The pipeline has five steps:

  1. An LLM identifies the key objects in the response.
  2. It writes questions about their existence, number and attributes.
  3. Expert vision models, an open-set detector and a VQA model, answer those questions.
  4. The answers are turned into a structured visual knowledge base of object- and attribute-level claims.
  5. An LLM rewrites the response guided by that knowledge base and attaches bounding boxes as evidence.

Because every intermediate output is visible, the process is interpretable, and it works as a plug-in for different MLLMs. Experiments on POPE, MME and LLaVA-QA90 show large reductions in hallucination. This suggests correction is a cheap, practical alternative to retraining.

Key points

  • Woodpecker is presented as the first correction-based, training-free approach to visual hallucination in MLLMs. It handles both object-level and attribute-level errors.
  • On POPE, accuracy rises from 54.67% to 85.33% for MiniGPT-4 and from 62% to 86.33% for mPLUG-Owl (gains of 30.66 and 24.33 points).
  • In ablations, the open-set detector accounts for most gains on existence and count questions, while the VQA model drives improvements on color.
  • In a correction analysis on MME using an always-"Yes" baseline, Woodpecker reached 79.2% accuracy with low omission and mis-correction rates.
  • Evaluation also included GPT-4V-based assessment, and the code is public at github.com/BradyFU/Woodpecker.

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

 Institutions

🏢TencentShenzhen · China163 works · 21,032 citations🎓University of Science and Technology of ChinaHefei · China145 works · 114,409 citations🏛️Chinese Academy of SciencesBeijing · China246 works · 24,951 citations🔬Institute of AutomationBeijing · China87 works · 14,883 citations🎓Nanjing UniversityNanjing · China99 works · 8,512 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksComputer Vision and Pattern Recognition · 3,486 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,894 worksVision & language · 811 works

Cite

Cite as Shukang Yin, Chaoyou Fu, Sirui Zhao, Tong Xu, Hao Wang, Dianbo Sui, Yunhang Shen, Ke Li, Xing Sun and Enhong Chen. 2023. Woodpecker: Hallucination Correction for Multimodal Large Language Models. arXiv:2310.16045 · SCIENCE CHINA Information Sciences, 2024.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1007/s11432-024-4251-x · arXiv 2310.16045

@misc{yin2023woodpecker,
  title = {{Woodpecker: Hallucination Correction for Multimodal Large Language Models}},
  author = {Shukang Yin, Chaoyou Fu, Sirui Zhao, Tong Xu, Hao Wang, Dianbo Sui, Yunhang Shen, Ke Li, Xing Sun and Enhong Chen},
  year = {2023},
  doi = {10.1007/s11432-024-4251-x},
  eprint = {2310.16045},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.1007/s11432-024-4251-x}
}

References

48 references, 28 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. Qwen-vl: A versatile vision-language model for understanding, localization, text reading, and beyond. arXiv, 2023. 1
  2. William Berrios, Gautam Mittal, Tristan Thrush, Douwe Kiela, and Amanpreet Singh. Towards language models that can see: Computer vision through the lens of natural language. arXiv, 2023. 3 find
  3. Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George Bm Van Den Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Clark, et al. Improving language models by retrieving from trillions of tokens. In ICML, 2022. 3 find
  4. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. NeurIPS, 2020. 5
  5. Emily Dinan, Stephen Roller, Kurt Shuster, Angela Fan, Michael Auli, and Jason Weston. Wizard of wikipedia: Knowledge-powered conversational agents. arXiv, 2018. 3
  6. Chaoyou Fu, Peixian Chen, Yunhang Shen, Yulei Qin, Mengdan Zhang, Xu Lin, Zhenyu Qiu, Wei Lin, Jinrui Yang, Xiawu Zheng, et al. Mme: A comprehensive evaluation benchmark for multimodal large language models. arXiv, 2023. 3, 5
  7. Ran Gong, Qiuyuan Huang, Xiaojian Ma, Hoi Vo, Zane Durante, Yusuke Noda, Zilong Zheng, Song-Chun Zhu, Demetri Terzopoulos, Li Fei-Fei, et al. Mindagent: Emergent gaming interaction. arXiv, 2023. 3 find
  8. Anisha Gunjal, Jihan Yin, and Erhan Bas. Detecting and preventing hallucinations in large vision language models. arXiv, 2023. 3
  9. Tanmay Gupta and Aniruddha Kembhavi. Visual programming: Compositional visual reasoning without training. In CVPR, 2023. 3
  10. Kung-Hsiang Huang, Hou Pong Chan, and Heng Ji. Zero-shot faithful factual error correction. arXiv, 2023. 3, 5 find
  11. Andrew Jaegle, Felix Gimeno, Andy Brock, Oriol Vinyals, Andrew Zisserman, and Joao Carreira. Perceiver: General perception with iterative attention. In ICML, 2021. 5
  12. Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C. Berg, Wan-Yen Lo, Piotr Dollár, and Ross Girshick. Segment anything. arXiv, 2023. 4
  13. Xin Lai, Zhuotao Tian, Yukang Chen, Yanwei Li, Yuhui Yuan, Shu Liu, and Jiaya Jia. Lisa: Reasoning segmentation via large language model. arXiv, 2023. 3 find
  14. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. NeurIPS, 2020. 3
  15. Bo Li, Yuanhan Zhang, Liangyu Chen, Jinghao Wang, Jingkang Yang, and Ziwei Liu. Otter: A multi-modal model with in-context instruction tuning. arXiv, 2023. 5, 6, 7
  16. Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. arXiv, 2023. 4, 5
  17. Liunian Harold Li, Pengchuan Zhang, Haotian Zhang, Jianwei Yang, Chunyuan Li, Yiwu Zhong, Lijuan Wang, Lu Yuan, Lei Zhang, Jenq-Neng Hwang, et al. Grounded language-image pre-training. In CVPR, 2022. 4 find
  18. Yifan Li, Yifan Du, Kun Zhou, Jinpeng Wang, Wayne Xin Zhao, and Ji-Rong Wen. Evaluating object hallucination in large vision-language models. arXiv, 2023. 3, 5, 6
  19. Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In ECCV, 2014. 5
  20. Fuxiao Liu, Kevin Lin, Linjie Li, Jianfeng Wang, Yaser Yacoob, and Lijuan Wang. Mitigating hallucination in large multi-modal models via robust instruction tuning. arXiv, 2023. 1, 3, 7
  21. Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. arXiv, 2023. 1, 3, 5, 6, 7
  22. Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. arXiv, 2023. 4, 5
  23. Jiaying Lu, Jinmeng Rao, Kezhen Chen, Xiaoyuan Guo, Yawen Zhang, Baochen Sun, Carl Yang, and Jie Yang. Evaluation and mitigation of agnosia in multimodal large language models. arXiv, 2023. 3 find
  24. Pan Lu, Baolin Peng, Hao Cheng, Michel Galley, Kai-Wei Chang, Ying Nian Wu, Song-Chun Zhu, and Jianfeng Gao. Chameleon: Plug-and-play compositional reasoning with large language models. arXiv, 2023. 3
  25. Timo Lüddecke and Alexander Ecker. Image segmentation using text and image prompts. In CVPR, 2022. 4 find
  26. OpenAI. Gpt-4 technical report. arXiv, 2023. 5
  27. Baolin Peng, Michel Galley, Pengcheng He, Hao Cheng, Yujia Xie, Yu Hu, Qiuyuan Huang, Lars Liden, Zhou Yu, Weizhu Chen, et al. Check your facts and try again: Improving large language models with external knowledge and automated feedback. arXiv, 2023. 3
  28. Fabio Petroni, Aleksandra Piktus, Angela Fan, Patrick Lewis, Majid Yazdani, Nicola De Cao, James Thorne, Yacine Jernite, Vladimir Karpukhin, Jean Maillard, et al. Kilt: a benchmark for knowledge intensive language tasks. arXiv, 2020. 3
  29. Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Dmytro Okhonko, Samuel Broscheit, Gautier Izacard, Patrick Lewis, Barlas Oğuz, Edouard Grave, Wen-tau Yih, et al. The web is your oyster-knowledge-intensive nlp against a very large web corpus. arXiv, 2021. 3 find
  30. Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, and Yueting Zhuang. Hugginggpt: Solving ai tasks with chatgpt and its friends in huggingface. arXiv, 2023. 3 find
  31. Kurt Shuster, Jing Xu, Mojtaba Komeili, Da Ju, Eric Michael Smith, Stephen Roller, Megan Ung, Moya Chen, Kushal Arora, Joshua Lane, et al. Blenderbot 3: a deployed conversational agent that continually learns to responsibly engage. arXiv, 2022. 3
  32. Amanpreet Singh, Ronghang Hu, Vedanuj Goswami, Guillaume Couairon, Wojciech Galuba, Marcus Rohrbach, and Douwe Kiela. Flava: A foundational language and vision alignment model. In CVPR, 2022. 4 find
  33. Bin Wang, Fan Wu, Xiao Han, Jiahui Peng, Huaping Zhong, Pan Zhang, Xiaoyi Dong, Weijia Li, Wei Li, Jiaqi Wang, et al. Vigc: Visual instruction generation and correction. arXiv, 2023. 1, 2, 3 find
  34. Junyang Wang, Yiyang Zhou, Guohai Xu, Pengcheng Shi, Chenlin Zhao, Haiyang Xu, Qinghao Ye, Ming Yan, Ji Zhang, Jihua Zhu, et al. Evaluation and analysis of hallucination in large vision-language models. arXiv, 2023. 3 find
  35. Teng Wang, Jinrui Zhang, Junjie Fei, Yixiao Ge, Hao Zheng, Yunlong Tang, Zhe Li, Mingqi Gao, Shanshan Zhao, Ying Shan, et al. Caption anything: Interactive image description with diverse multimodal controls. arXiv, 2023. 3 find
  36. Zhenyu Wang, Yali Li, Xi Chen, Ser-Nam Lim, Antonio Torralba, Hengshuang Zhao, and Shengjin Wang. Detecting everything in the open world: Towards universal object detection. In CVPR, 2023. 4 find
  37. Chenfei Wu, Shengming Yin, Weizhen Qi, Xiaodong Wang, Zecheng Tang, and Nan Duan. Visual chatgpt: Talking, drawing and editing with visual foundation models. arXiv, 2023. 3
  38. Rui Yang, Lin Song, Yanwei Li, Sijie Zhao, Yixiao Ge, Xiu Li, and Ying Shan. Gpt4tools: Teaching large language model to use tools via self-instruction. arXiv, 2023. 3
  39. Zhengyuan Yang, Linjie Li, Jianfeng Wang, Kevin Lin, Ehsan Azarnasab, Faisal Ahmed, Zicheng Liu, Ce Liu, Michael Zeng, and Lijuan Wang. Mm-react: Prompting chatgpt for multimodal reasoning and action. arXiv, 2023. 3
  40. Qinghao Ye, Haiyang Xu, Guohai Xu, Jiabo Ye, Ming Yan, Yiyang Zhou, Junyang Wang, Anwen Hu, Pengcheng Shi, Yaya Shi, et al. mplug-owl: Modularization empowers large language models with multimodality. arXiv, 2023. 1, 3, 5, 6, 7
  41. Shukang Yin, Chaoyou Fu, Sirui Zhao, Ke Li, Xing Sun, Tong Xu, and Enhong Chen. A survey on multimodal large language models. arXiv, 2023. 1, 3, 5
  42. Haoxuan You, Rui Sun, Zhecan Wang, Long Chen, Gengyu Wang, Hammad A Ayyubi, Kai-Wei Chang, and Shih-Fu Chang. Idealgpt: Iteratively decomposing vision and language reasoning via large language models. arXiv, 2023. 3 find
  43. Andy Zeng, Maria Attarian, Brian Ichter, Krzysztof Choromanski, Adrian Wong, Stefan Welker, Federico Tombari, Aveek Purohit, Michael Ryoo, Vikas Sindhwani, et al. Socratic models: Composing zero-shot multimodal reasoning with language. arXiv, 2022. 3
  44. Ao Zhang, Hao Fei, Yuan Yao, Wei Ji, Li Li, Zhiyuan Liu, and Tat-Seng Chua. Transfer visual prompt generator across llms. arXiv, 2023. 1 find
  45. Renrui Zhang, Xiangfei Hu, Bohao Li, Siyuan Huang, Hanqiu Deng, Yu Qiao, Peng Gao, and Hongsheng Li. Prompt, generate, then cache: Cascade of foundation models makes strong few-shot learners. In CVPR, 2023. 3 find
  46. Deyao Zhu, Jun Chen, Kilichbek Haydarov, Xiaoqian Shen, Wenxuan Zhang, and Mohamed Elhoseiny. Chatgpt asks, blip-2 answers: Automatic questioning towards enriched visual descriptions. arXiv, 2023. 3 find
  47. Deyao Zhu, Jun Chen, Xiaoqian Shen, Xiang Li, and Mohamed Elhoseiny. Minigpt-4: Enhancing vision-language understanding with advanced large language models. arXiv, 2023. 1, 3, 5, 6, 7
  48. Xiangyang Zhu, Renrui Zhang, Bowei He, Ziyu Guo, Ziyao Zeng, Zipeng Qin, Shanghang Zhang, and Peng Gao. Pointclip v2: Prompting clip and gpt for powerful 3d open-world learning. In ICCV, 2023. 3 find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yin-2023-woodpecker-hallucination-correction-multimodal-large/