![[2412.00686] LVLM-COUNT: Enhancing the Counting Ability of Large Vision-Language Models](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2412.00686] LVLM-COUNT: Enhancing the Counting Ability of Large Vision-Language Models
Summary
The paper presents LVLM-COUNT, a method to enhance the counting ability of large vision-language models (LVLMs) by using a divide-and-conquer approach to improve performance on visual counting tasks.
Why It Matters
Counting is essential for various visual tasks, and LVLMs currently struggle with larger object counts. This research addresses a significant gap in their capabilities, providing a new method that could improve applications in computer vision and AI, making it relevant for developers and researchers in the field.
Key Takeaways
- LVLMs perform well with small object counts but struggle with larger numbers.
- The proposed divide-and-conquer method effectively enhances counting accuracy.
- The approach prevents repetitive counting, a common issue in naive implementations.
- The method is validated across various datasets and benchmarks.
- This research serves as a reference for future improvements in LVLM capabilities.
Computer Science > Computer Vision and Pattern Recognition arXiv:2412.00686 (cs) [Submitted on 1 Dec 2024 (v1), last revised 16 Feb 2026 (this version, v4)] Title:LVLM-COUNT: Enhancing the Counting Ability of Large Vision-Language Models Authors:Muhammad Fetrat Qharabagh, Mohammadreza Ghofrani, Kimon Fountoulakis View a PDF of the paper titled LVLM-COUNT: Enhancing the Counting Ability of Large Vision-Language Models, by Muhammad Fetrat Qharabagh and 2 other authors View PDF HTML (experimental) Abstract:Counting is a fundamental operation for various real-world visual tasks, requiring both object recognition and robust counting capabilities. Despite their advanced visual perception, large vision-language models (LVLMs) are known to struggle with counting tasks. In this work, we evaluate the performance of several LVLMs on visual counting tasks across multiple counting and vision datasets. We observe that while their performance may be less prone to error for small numbers of objects, they exhibit significant weaknesses as the number of objects increases. To alleviate this issue, we propose a simple yet effective baseline method that enhances LVLMs' counting ability for large numbers of objects using a divide-and-conquer approach. Our method decomposes counting problems into sub-tasks. Moreover, it incorporates a mechanism to prevent objects from being split during division, which could otherwise lead to repetitive counting -- a common issue in a naive divide-and-conquer im...