UNICBench: UNIfied Counting Benchmark for MLLM

TL;DR

UNICBench is a comprehensive benchmark for evaluating counting abilities of multimodal large language models across image, text, and audio, revealing strengths and gaps in current models and providing a standardized evaluation toolkit.

Contribution

The paper introduces UNICBench, a unified, multi-modal counting benchmark with detailed evaluation protocols and a large annotated dataset for rigorous assessment of MLLMs.

Findings

Strong performance on basic counting tasks

Significant gaps in reasoning and complex partitions

Identifies long-tail errors and room for improvement

Abstract

Counting is a core capability for multimodal large language models (MLLMs), yet there is no unified counting dataset to rigorously evaluate this ability across image, text, and audio. We present UNICBench, a unified multimodal, multi level counting benchmark and evaluation toolkit with accurate ground truth, deterministic numeric parsing, and stratified reporting. The corpus comprises 5,300 images (5,508 QA), 872 documents (5,888 QA), and 2,069 audio clips (2,905 QA), annotated with a three level capability taxonomy and difficulty tags. Under a standardized protocol with fixed splits/prompts/seeds and modality specific matching rules, we evaluate 45 state-of-the-art MLLMs across modalities. Results show strong performance on some basic counting tasks but significant gaps on reasoning and the hardest partitions, highlighting long-tail errors and substantial headroom for improving general counting. UNICBench offers a rigorous and comparable basis for measurement and a public toolkit to accelerate progress.