Counting to Four is still a Chore for VLMs

TL;DR

This paper investigates why vision-language models struggle with simple counting tasks, revealing that visual evidence is underused during reasoning and proposing interventions to improve counting accuracy.

Contribution

It introduces COUNTINGTRICKS, a controlled evaluation suite, and analyzes model behavior, highlighting the importance of visual evidence during language reasoning.

Findings

Visual evidence is strongest in the modality projection stage.

Counting failures are due to visual perception limits and underuse of visual evidence in language reasoning.

Modality Attention Share improves counting performance.

Abstract

Vision--language models (VLMs) have achieved impressive performance on complex multimodal reasoning tasks, yet they still fail on simple grounding skills such as object counting. Existing evaluations mostly assess only final outputs, offering limited insight into where these failures arise inside the model. In this work, we present an empirical study of VLM counting behavior through both behavioral and mechanistic analysis. We introduce COUNTINGTRICKS, a controlled evaluation suite of simple shape-based counting cases designed to expose vulnerabilities under different patchification layouts and adversarial prompting conditions. Using attention analysis and component-wise probing, we show that count-relevant visual evidence is strongest in the modality projection stage but degrades substantially in later language layers, where models become more susceptible to text priors. Motivated by this finding, we further evaluate Modality Attention Share (MAS), a lightweight intervention that encourages a minimum budget of visual attention during answer generation. Our results suggest that counting failures in VLMs stem not only from visual perception limits, but also from the underuse of visual evidence during language-stage reasoning. Code and dataset will be released at https://github.com/leduy99/-CVPRW26-Modality-Attention-Share.