# Collage Prompting: Budget-Friendly Visual Recognition with GPT-4V

**Authors:** Siyu Xu, Yunke Wang, Daochang Liu, Chang Xu  
**arXiv:** [2403.11468](https://arxiv.org/abs/2403.11468)  
**Date:** 2024-03-18  
**Summary page:** https://www.summarizepaper.com/en/paper/2403.11468v1/

## Key points

- Authors: Siyu Xu, Yunke Wang, Daochang Liu, Chang Xu
- Paper title: "Collage Prompting: Budget-Friendly Visual Recognition with GPT-4V"
- Collage Prompting proposed as a cost-effective approach for GPT-4V in image recognition tasks
- Combining multiple images into a single visual input enhances GPT-4V efficiency
- Accuracy of image recognition influenced by order of images in collage prompt
- Method focuses on optimizing image arrangement for maximum accuracy
- Graph predictor trained to assess accuracy of each collage prompt
- Optimization method proposed to navigate through possible image arrangements
- Experimental results show Collage Prompting outperforms standard methods in cost-efficiency and accuracy
- Importance of thoughtful arrangement emphasized for maximizing AI performance

## Summary

In their paper titled "Collage Prompting: Budget-Friendly Visual Recognition with GPT-4V," authors Siyu Xu, Yunke Wang, Daochang Liu, and Chang Xu delve into the realm of generative AI advancements and the potential of GPT-4V in image recognition tasks. The authors propose Collage Prompting as a cost-effective approach to overcome the financial barrier associated with GPT-4V's inference process. By combining multiple images into a single visual input for GPT-4V, Collage Prompting enhances its efficiency in image recognition tasks. The accuracy of GPT-4V's image recognition is heavily influenced by the order of images within the collage prompt. To address this variability, their method focuses on optimizing the arrangement of images to maximize recognition accuracy. A graph predictor is trained to assess the accuracy of each collage prompt and an optimization method is proposed to navigate through possible image arrangements. Experimental results demonstrate that Collage Prompting outperforms standard prompting methods in terms of cost-efficiency and accuracy when utilizing learned image arrangements within collage prompts. This innovative approach highlights the importance of thoughtful arrangement in maximizing AI performance in image recognition tasks.