Elamur Research
☰
home / papers / 2023 / Improved Baselines with Visual Instruction Tuning

Improved Baselines with Visual Instruction Tuning

Written for the Research library (claude-opus-5-5) from the paper text. Original — arxiv.org/pdf/2310.03744

Summary

Large multimodal models (LMMs) differ widely in architecture and training data, so it has been hard to tell which design choices actually matter. For example, LLaVA does well in open-ended visual chat, while InstructBLIP is stronger on short-answer VQA. The authors run a controlled study within the LLaVA framework and find that its plain fully-connected vision-language connector is surprisingly strong and data-efficient. Their improved baseline, LLaVA-1.5, makes three simple changes:

  • it uses a CLIP-ViT-L-336px encoder,
  • it replaces the linear projection with an MLP,
  • it adds academic VQA data, with a brief instruction appended to each question to request short answers.

The format instruction lets one model switch between short and detailed replies. LLaVA-1.5 reaches state-of-the-art results on 11 benchmarks using only public data. The paper also explores higher-resolution inputs, compositional generalization, data efficiency and hallucination. Because training is cheap and reproducible, the work lowers the barrier for open LMM research.

Key points

  • The final 13B checkpoint is trained on about 1.2M publicly available samples and takes roughly one day on a single 8-A100 node. InstructBLIP and Qwen-VL, by contrast, pretrain resamplers on 129M and 1.4B image-text pairs, respectively.
  • Adding VQAv2 together with the response formatting prompt raises the MME score from 809.6 to 1323.8, which is 111 points above InstructBLIP.
  • Randomly cutting the training mixture by up to 75% causes no significant performance drop.
  • Splitting images into grids allows higher-resolution input, which improves fine-grained perception and reduces hallucination.
  • Training on English visual data plus multilingual text-only ShareGPT yields multilingual visual conversation, though errors remain (e.g., in Korean). Single-image-only input and remaining hallucinations are acknowledged as limitations.

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

Fields: Other

Cited by

920 papers of our library cite it (of the 140673 whose bibliographies we read); 34 of them have a page here, 886 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2310.03744 · arXiv 2310.03744

@misc{liu2023improved,
  title = {{Improved Baselines with Visual Instruction Tuning}},
  author = {Liu, Haotian and Li, Chunyuan and Li, Yuheng and Lee, Yong Jae},
  year = {2023},
  doi = {10.48550/arxiv.2310.03744},
  eprint = {2310.03744},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2310.03744}
}

References

62 references, 23 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Adept AI. Fuyu-8b: A multimodal architecture for ai agents. https://www.adept.ai/blog/fuyu-8b, 2024. 2 ↗
  2. Jean-Baptiste Alayrac, Jeff Donahue, Pauline Luc, Antoine Miech, Iain Barr, Yana Hasson, Karel Lenc, Arthur Mensch, Katie Millican, Malcolm Reynolds, et al. Flamingo: a visual language model for few-shot learning. arXiv preprint arXiv:2204.14198, 2022. 1 find our card
  3. 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. arXiv preprint arXiv:2308.12966, 2023. 1, 2, 4, 5, 6, 13 find our card
  4. Yonatan Bitton, Hritik Bansal, Jack Hessel, Rulin Shao, Wanrong Zhu, Anas Awadalla, Josh Gardner, Rohan Taori, and Ludwig Schimdt. Visit-bench: A benchmark for vision-language instruction following inspired by real-world use, 2023. 3 find
  5. Kevin Black, Michael Janner, Yilun Du, Ilya Kostrikov, and Sergey Levine. Training diffusion models with reinforcement learning. arXiv preprint arXiv:2305.13301, 2023. 1 — [arXiv:2305.13301](https://arxiv.org/abs/2305.13301) find
  6. Nicholas Carlini, Milad Nasr, Christopher A Choquette-Choo, Matthew Jagielski, Irena Gao, Anas Awadalla, Pang Wei Koh, Daphne Ippolito, Katherine Lee, Florian Tramer, et al. Are aligned neural networks adversarially aligned? arXiv preprint arXiv:2306.15447, 2023. 1 find
  7. Delong Chen, Jianfeng Liu, Wenliang Dai, and Baoyuan Wang. Visual instruction tuning with polite flamingo. arXiv preprint arXiv:2307.01003, 2023. 2, 3 — [arXiv:2307.01003](https://arxiv.org/abs/2307.01003) find
  8. Keqin Chen, Zhao Zhang, Weili Zeng, Richong Zhang, Feng Zhu, and Rui Zhao. Shikra: Unleashing multimodal llm’s referential dialogue magic. arXiv preprint arXiv:2306.15195, 2023. 1, 2, 4, 5 find our card
  9. Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. In ICML, 2020. 3 find our card
  10. Xinlei Chen, Haoqi Fan, Ross Girshick, and Kaiming He. Improved baselines with momentum contrastive learning. arXiv preprint arXiv:2003.04297, 2020. 3 — [arXiv:2003.04297](https://arxiv.org/abs/2003.04297) find
  11. Xi Chen, Josip Djolonga, Piotr Padlewski, Basil Mustafa, Soravit Changpinyo, Jialin Wu, Carlos Riquelme Ruiz, Sebastian Goodman, Xiao Wang, Yi Tay, et al. Pali-x: On scaling up a multilingual vision and language model. arXiv preprint arXiv:2305.18565, 2023. 5 — [arXiv:2305.18565](https://arxiv.org/abs/2305.18565) find
  12. Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. Vicuna: An open-source chatbot impressing gpt-4 with 90%* chatgpt quality, 2023. 9 find
  13. Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, et al. Scaling instruction-finetuned language models. arXiv preprint arXiv:2210.11416, 2022. 2 find
  14. Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, and Steven Hoi. Instructblip: Towards general-purpose vision-language models with instruction tuning. arXiv preprint arXiv:2305.06500, 2023. 1, 2, 3, 5, 6, 8, 13 find our card
  15. Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. 4 find
  16. Yuxin Fang, Wen Wang, Binhui Xie, Quan Sun, Ledell Wu, Xinggang Wang, Tiejun Huang, Xinlong Wang, and Yue Cao. Eva: Exploring the limits of masked visual representation learning at scale. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19358– 19369, 2023.… find
  17. 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 preprint arXiv:2306.13394, 2023. 1, 4, 5 find our card
  18. Tao Gong, Chengqi Lyu, Shilong Zhang, Yudong Wang, Miao Zheng, Qian Zhao, Kuikun Liu, Wenwei Zhang, Ping Luo, and Kai Chen. Multimodal-gpt: A vision and language model for dialogue with humans. arXiv preprint arXiv:2305.04790, 2023. 1 — [arXiv:2305.04790](https://arxiv.org/abs/2305.04790) find
  19. Yash Goyal, Tejas Khot, Douglas Summers-Stay, Dhruv Batra, and Devi Parikh. Making the v in vqa matter: Elevating the role of image understanding in visual question answering. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 6904–6913, 2017. 3, 5, 6, 9 find our card
  20. Danna Gurari, Qing Li, Abigale J Stangl, Anhong Guo, Chi Lin, Kristen Grauman, Jiebo Luo, and Jeffrey P Bigham. Vizwiz grand challenge: Answering visual questions from blind people. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3608–3617, 2018. 5, 6 find
  21. Drew A Hudson and Christopher D Manning. Gqa: A new dataset for real-world visual reasoning and compositional question answering. In CVPR, 2019. 4, 5, 9 find our card
  22. IDEFICS. Introducing idefics: An open reproduction of state-of-the-art visual language model. https : / / huggingface.co/blog/idefics, 2023. 5, 6 find
  23. Gabriel Ilharco, Mitchell Wortsman, Ross Wightman, Cade Gordon, Nicholas Carlini, Rohan Taori, Achal Dave, Vaishaal Shankar, Hongseok Namkoong, John Miller, Hannaneh Hajishirzi, Ali Farhadi, and Ludwig Schmidt. Openclip. 2021. If you use this software, please cite it as below. 6 find
  24. Sahar Kazemzadeh, Vicente Ordonez, Mark Matten, and Tamara Berg. Referitgame: Referring to objects in photographs of natural scenes. In Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pages 787–798, 2014. 4, 9 find
  25. Ranjay Krishna, Yuke Zhu, Oliver Groth, Justin Johnson, Kenji Hata, Joshua Kravitz, Stephanie Chen, Yannis Kalantidis, Li-Jia Li, David A Shamma, et al. Visual genome: Connecting language and vision using crowdsourced dense image annotations. International journal of computer vision, 123:32–73, 2017… find our card
  26. Xin Lai, Zhuotao Tian, Yukang Chen, Yanwei Li, Yuhui Yuan, Shu Liu, and Jiaya Jia. Lisa: Reasoning segmentation via large language model. arXiv preprint arXiv:2308.00692, 2023. 1 — [arXiv:2308.00692](https://arxiv.org/abs/2308.00692) find
  27. Bohao Li, Rui Wang, Guangzhi Wang, Yuying Ge, Yixiao Ge, and Ying Shan. Seed-bench: Benchmarking multimodal llms with generative comprehension. arXiv preprint arXiv:2307.16125, 2023. 1, 5 find our card
  28. Bo Li, Peiyuan Zhang, Jingkang Yang, Yuanhan Zhang, Fanyi Pu, and Ziwei Liu. Otterhd: A high-resolution multi-modality model, 2023. 2 find
  29. Bo Li, Yuanhan Zhang, Liangyu Chen, Jinghao Wang, Fanyi Pu, Jingkang Yang, Chunyuan Li, and Ziwei Liu. Mimic-it: Multi-modal in-context instruction tuning. arXiv preprint arXiv:2306.05425, 2023. 1 — [arXiv:2306.05425](https://arxiv.org/abs/2306.05425) find
  30. Chunyuan Li, Zhe Gan, Zhengyuan Yang, Jianwei Yang, Linjie Li, Lijuan Wang, and Jianfeng Gao. Multimodal foundation models: From specialists to general-purpose assistants. arXiv preprint arXiv:2309.10020, 2023. 1 — [arXiv:2309.10020](https://arxiv.org/abs/2309.10020) find
  31. Chunyuan Li, Cliff Wong, Sheng Zhang, Naoto Usuyama, Haotian Liu, Jianwei Yang, Tristan Naumann, Hoifung Poon, and Jianfeng Gao. Llava-med: Training a large language-and-vision assistant for biomedicine in one day. arXiv preprint arXiv:2306.00890, 2023. 1 find our card
  32. 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 preprint arXiv:2301.12597, 2023. 1, 2, 3, 4, 5, 6, 13 find our card
  33. Xiang Lisa Li and Percy Liang. Prefix-tuning: Optimizing continuous prompts for generation. arXiv preprint arXiv:2101.00190, 2021. 3 find
  34. Yifan Li, Yifan Du, Kun Zhou, Jinpeng Wang, Wayne Xin Zhao, and Ji-Rong Wen. Evaluating object hallucination in large vision-language models. arXiv preprint arXiv:2305.10355, 2023. 1, 5 find our card
  35. 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. 2, 5 find our card
  36. Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. In NeurIPS, 2023. 1, 2, 3, 4, 5, 6, 8, 9 find our card
  37. Yuan Liu, Haodong Duan, Yuanhan Zhang, Bo Li, Songyang Zhang, Wangbo Zhao, Yike Yuan, Jiaqi Wang, Conghui He, Ziwei Liu, et al. Mmbench: Is your multi-modal model an all-around player? arXiv preprint arXiv:2307.06281, 2023. 1, 5, 7 find our card
  38. Pan Lu, Swaroop Mishra, Tanglin Xia, Liang Qiu, Kai-Wei Chang, Song-Chun Zhu, Oyvind Tafjord, Peter Clark, and Ashwin Kalyan. Learn to explain: Multimodal reasoning via thought chains for science question answering. Advances in Neural Information Processing Systems, 2022. 5 find
  39. Yadong Lu, Chunyuan Li, Haotian Liu, Jianwei Yang, Jianfeng Gao, and Yelong Shen. An empirical study of scaling instruct-tuned large multimodal models. arXiv preprint arXiv:2309.09958, 2023. 1, 4 — [arXiv:2309.09958](https://arxiv.org/abs/2309.09958) find
  40. Junhua Mao, Jonathan Huang, Alexander Toshev, Oana Camburu, Alan L Yuille, and Kevin Murphy. Generation and comprehension of unambiguous object descriptions. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 11–20, 2016. 4, 9 find our card
  41. Kenneth Marino, Mohammad Rastegari, Ali Farhadi, and Roozbeh Mottaghi. Ok-vqa: A visual question answering benchmark requiring external knowledge. In Conference on Computer Vision and Pattern Recognition (CVPR), 2019. 4, 9 find
  42. Anand Mishra, Shashank Shekhar, Ajeet Kumar Singh, and Anirban Chakraborty. Ocr-vqa: Visual question answering by reading text in images. In 2019 international conference on document analysis and recognition (ICDAR), pages 947–952. IEEE, 2019. 4, 9 find
  43. OpenAI. Gpt-4v(ision) system card. https://cdn. openai .com /papers/GPTV\_System\_Card.pdf, 2023. 1 find
  44. Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. arXiv preprint arXiv:2103.00020, 2021. 6 find our card
  45. Dustin Schwenk, Apoorv Khandelwal, Christopher Clark, Kenneth Marino, and Roozbeh Mottaghi. A-okvqa: A benchmark for visual question answering using world knowledge. In European Conference on Computer Vision, pages 146–162. Springer, 2022. 4, 9 find our card
  46. ShareGPT. https://sharegpt.com/, 2023. 4, 7, 8, 9 ↗
  47. Oleksii Sidorov, Ronghang Hu, Marcus Rohrbach, and Amanpreet Singh. Textcaps: a dataset for image captioning with reading comprehension. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part II 16, pages 742–758. Springer, 2020. 4, 9 find
  48. Amanpreet Singh, Vivek Natarajan, Meet Shah, Yu Jiang, Xinlei Chen, Dhruv Batra, Devi Parikh, and Marcus Rohrbach. Towards vqa models that can read. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8317–8326, 2019. 5 find
  49. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288, 2023. 7 find
  50. Wenhai Wang, Zhe Chen, Xiaokang Chen, Jiannan Wu, Xizhou Zhu, Gang Zeng, Ping Luo, Tong Lu, Jie Zhou, Yu Qiao, et al. Visionllm: Large language model is also an open-ended decoder for vision-centric tasks. arXiv preprint arXiv:2305.11175, 2023. 1 — [arXiv:2305.11175](https://arxiv.org/abs/2305.11175) find
  51. Jason Wei, Maarten Bosma, Vincent Y Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M Dai, and Quoc V Le. Finetuned language models are zero-shot learners. arXiv preprint arXiv:2109.01652, 2021. 2 find
  52. Zhengyuan Yang, Linjie Li, Kevin Lin, Jianfeng Wang, Chung-Ching Lin, Zicheng Liu, and Lijuan Wang. The dawn of lmms: Preliminary explorations with gpt-4v (ision). arXiv preprint arXiv:2309.17421, 2023. 6 find our card
  53. Jiabo Ye, Anwen Hu, Haiyang Xu, Qinghao Ye, Ming Yan, Guohai Xu, Chenliang Li, Junfeng Tian, Qi Qian, Ji Zhang, Qin Jin, Liang He, Xin Alex Lin, and Fei Huang. Ureader: Universal ocr-free visually-situated language understanding with multimodal large language model, 2023. 2 find
  54. 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 preprint arXiv:2304.14178, 2023. 1 find our card
  55. Weihao Yu, Zhengyuan Yang, Linjie Li, Jianfeng Wang, Kevin Lin, Zicheng Liu, Xinchao Wang, and Lijuan Wang. Mm-vet: Evaluating large multimodal models for integrated capabilities. arXiv preprint arXiv:2308.02490, 2023. 1, 3, 4, 5, 8 find our card
  56. Shilong Zhang, Peize Sun, Shoufa Chen, Min Xiao, Wenqi Shao, Wenwei Zhang, Kai Chen, and Ping Luo. Gpt4roi: Instruction tuning large language model on region-of-interest. arXiv preprint arXiv:2307.03601, 2023. 1 — [arXiv:2307.03601](https://arxiv.org/abs/2307.03601) find
  57. Yanzhe Zhang, Ruiyi Zhang, Jiuxiang Gu, Yufan Zhou, Nedim Lipka, Diyi Yang, and Tong Sun. Llavar: Enhanced visual instruction tuning for text-rich image understanding. arXiv preprint arXiv:2306.17107, 2023. 2 find
  58. Bo Zhao, Boya Wu, and Tiejun Huang. Svit: Scaling up visual instruction tuning. arXiv preprint arXiv:2307.04087, 2023. 1, 2 — [arXiv:2307.04087](https://arxiv.org/abs/2307.04087) find
  59. Yunqing Zhao, Tianyu Pang, Chao Du, Xiao Yang, Chongxuan Li, Ngai-Man Cheung, and Min Lin. On evaluating adversarial robustness of large vision-language models. arXiv preprint arXiv:2305.16934, 2023. 1 — [arXiv:2305.16934](https://arxiv.org/abs/2305.16934) find
  60. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric. P Xing, Hao Zhang, Joseph E. Gonzalez, and Ion Stoica. Judging llm-as-a-judge with mt-bench and chatbot arena, 2023. 10 find
  61. Chunting Zhou, Pengfei Liu, Puxin Xu, Srini Iyer, Jiao Sun, Yuning Mao, Xuezhe Ma, Avia Efrat, Ping Yu, Lili Yu, et al. Lima: Less is more for alignment. arXiv preprint arXiv:2305.11206, 2023. 2, 8 find
  62. Deyao Zhu, Jun Chen, Xiaoqian Shen, Xiang Li, and Mohamed Elhoseiny. Minigpt-4: Enhancing vision-language understanding with advanced large language models. arXiv preprint arXiv:2304.10592, 2023. 1, 2 find our card