CoCoT: Contrastive Chain-of-Thought Prompting for Large Multimodal Models with Multiple Image Inputs
When exploring the development of Artificial General Intelligence (AGI), a critical task for these models involves interpreting and processing information from multiple image inputs. However, Large Multimodal Models (LMMs) encounter two issues in such scenarios: (1) a lack of fine-grained perception, and (2) a tendency to blend information across multiple images. We first extensively investigate the capability of LMMs to perceive fine-grained visual details when dealing with multiple input images. The research focuses on two aspects: first, image-to-image matching (to evaluate whether LMMs can effectively reason and pair relevant images), and second, multi-image-to-text matching (to assess whether LMMs can accurately capture and summarize detailed image information). We conduct evaluations on a range of both open-source and closed-source large models, including GPT-4V, Gemini, OpenFlamingo, and MMICL. To enhance model performance, we further develop a Contrastive Chain-of-Thought (CoCoT) prompting approach based on multi-input multimodal models. This method requires LMMs to compare the similarities and differences among multiple image inputs, and then guide the models to answer detailed questions about multi-image inputs based on the identified similarities and differences. Our experimental results showcase CoCoT's proficiency in enhancing the multi-image comprehension capabilities of large multimodal models.
PDF Abstract
Datasets
RAVEN
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Visual Reasoning | Winoground | GPT-4V + CoCoT | Text Score | 58.5 | # 4 | |
| Image Score | 49.5 | # 4 | ||||
| Group Score | 44.5 | # 4 | ||||
| Visual Reasoning | Winoground | GPT-4V | Text Score | 54.5 | # 6 | |
| Image Score | 42.5 | # 10 | ||||
| Group Score | 37.75 | # 10 | ||||
| Visual Reasoning | Winoground | Gemini + CoCoT | Text Score | 40 | # 36 | |
| Image Score | 32.5 | # 19 | ||||
| Group Score | 27.75 | # 16 | ||||
| Visual Reasoning | Winoground | Gemini + CCoT | Text Score | 22.5 | # 95 | |
| Image Score | 33 | # 18 | ||||
| Group Score | 20.75 | # 28 | ||||
| Visual Reasoning | Winoground | Gemini + DDCoT | Text Score | 45 | # 17 | |
| Image Score | 25 | # 34 | ||||
| Group Score | 23.75 | # 19 | ||||
| Visual Reasoning | Winoground | Gemini | Text Score | 30.75 | # 62 | |
| Image Score | 26 | # 29 | ||||
| Group Score | 25 | # 18 | ||||
| Visual Reasoning | Winoground | MMICL + CoCoT | Text Score | 64.25 | # 3 | |
| Image Score | 52.5 | # 3 | ||||
| Group Score | 50.75 | # 2 | ||||
| Visual Reasoning | Winoground | MMICL + CCoT | Text Score | 51 | # 9 | |
| Image Score | 48 | # 5 | ||||
| Group Score | 47.5 | # 3 | ||||
| Visual Reasoning | Winoground | MMICL + DDCoT | Text Score | 46.75 | # 12 | |
| Image Score | 45 | # 8 | ||||
| Group Score | 36.75 | # 11 | ||||
| Visual Reasoning | Winoground | OpenFlamingo + CoCoT | Text Score | 58.25 | # 5 | |
| Image Score | 55.25 | # 2 | ||||
| Group Score | 41.5 | # 6 | ||||
| Visual Reasoning | Winoground | OpenFlamingo + CCoT | Text Score | 42.5 | # 27 | |
| Image Score | 27.5 | # 24 | ||||
| Group Score | 20 | # 31 | ||||
| Visual Reasoning | Winoground | OpenFlamingo + DDCoT | Text Score | 47.5 | # 10 | |
| Image Score | 47.25 | # 6 | ||||
| Group Score | 39 | # 8 | ||||
| Visual Reasoning | Winoground | OpenFlamingo | Text Score | 39 | # 39 | |
| Image Score | 41.25 | # 13 | ||||
| Group Score | 33.25 | # 12 |
