HoloCount: A Holistic Visual Counting Benchmark for MLLMs

Abstract

Visual counting is a fundamental pillar of multimodal intelligence, requiring a seamless integration of fine-grained grounding and spatial reasoning. While Multimodal Large Language Models (MLLMs) have achieved remarkable success in qualitative scene understanding, their quantitative precision remains a significant bottleneck, often characterized by persistent numerical hallucinations. Existing counting benchmarks primarily focus on basic perception in simplified contexts, failing to capture the complex failure modes that emerge under logical constraints or adversarial conditions. To address these limitations, we introduce HoloCount, a holistic and diagnostically rich benchmark structured around a three-level hierarchical taxonomy. HoloCount evaluates MLLMs across: (1) Semantic Counting, focusing on atomic and property-based enumeration; (2) Analytical Counting, assessing logical composition through spatial and set-based reasoning; and (3) Robustness Testing, probing model integrity against adverse scenarios and grounded counter-priors, such as high-density scenes and linguistic biases. Through an exhaustive evaluation of over 20 state-of-the-art MLLMs, we reveal a critical performance gap: even top-tier models degrade significantly as tasks transition from perception to complex analytical reasoning and adverse scenarios. Our findings provide a systematic landscape of current MLLM counting capabilities and offer a roadmap for developing more grounded and reliable multimodal systems.

Overview

• 2,480 QA pairs
• 20 fine-grained subsets
• 1,481 visual concepts
• 30+ evaluated MLLMs

Taxonomy

The benchmark features a three-level hierarchical taxonomy:

• Semantic Counting — Atomic and property-based enumeration (6 subsets)
• Analytical Counting — Logical composition through spatial and set-based reasoning (7 subsets)
• Robustness Testing — Model integrity against adverse scenarios (7 subsets)

Dataset Composition

Split	Count
Action & State Attribution	102
Aerial Perspective Shift	109
Atomic Counting	182
Categorical Cardinality	101
Chromatic Attribution	202
Complementary Exclusion	101
Coordinate-Prompt Region Grounding	84
Differential Comparison	100
General Semantic Attribution	101
High-Density Enumeration	100
Joint-Set Aggregation	100
Linguistic Prior Conflict	163
Material Attribution	101
Null-Target Prompting	250
Partial Object Occlusion	103
Relative Canonical Orientation	78
Scale Attribution	100
Small-Scale Enumeration	101
Visual Distractor Illusion	200
Visual-Prompt Region Grounding	102

Usage

from datasets import load_dataset

dataset = load_dataset("MM-MVR/HoloCount")

Fields

• image: the image file
• question: counting question
• answer: ground truth count (string)
• split: capability category
• sample_id: unique identifier
• image_height: original image height in pixels
• image_width: original image width in pixels

Citation

@article{deng2026holocount,
  title={HoloCount: A Holistic Visual Counting Benchmark for MLLMs},
  author={Deng, Jinhong and Qiao, Limeng and Wan, Guanglu},
  year={2026}
}

License

This dataset is released under the CC BY 4.0 License.

Contact

• Jinhong Deng (dengjinhong@meituan.com)