Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

KidVis: Do Multimodal Large Language Models Possess the Visual Perceptual Capabilities of a 6-Year-Old?

Xianfeng Wang, Kaiwei Zhang, Qi Jia, Zijian Chen, Guangtao Zhai, Xiongkuo Min

TL;DR

KidVis systematically probes whether Multimodal LLMs possess foundational visual primitives by decomposing vision into six atomic capabilities and evaluating 20 MLLMs against a 6–7-year-old human baseline. The benchmark reveals a stark gap between human perceptual baselines and contemporary models, with proprietary systems lagging by roughly 28 points and open-source models performing often near chance on several tasks. A detailed analysis assigns failure modes to soft attention, architectural tokenization, semantic priors, and lack of grounded reference frames, while exposing a Scaling Law Paradox where increasing parameters fails to yield proportional perceptual gains. These findings suggest that achieving generalized visual intelligence requires architectural innovations and developmentally informed mechanisms beyond mere scale, emphasizing true physical perception grounding. The work provides a diagnostic framework and benchmarks to steer future research toward embodied-like visual perception and robust grounding in pixel-level reality.

Abstract

While Multimodal Large Language Models (MLLMs) have demonstrated impressive proficiency in high-level reasoning tasks, such as complex diagrammatic interpretation, it remains an open question whether they possess the fundamental visual primitives comparable to human intuition. To investigate this, we introduce KidVis, a novel benchmark grounded in the theory of human visual development. KidVis deconstructs visual intelligence into six atomic capabilities - Concentration, Tracking, Discrimination, Memory, Spatial, and Closure - already possessed by 6-7 year old children, comprising 10 categories of low-semantic-dependent visual tasks. Evaluating 20 state-of-the-art MLLMs against a human physiological baseline reveals a stark performance disparity. Results indicate that while human children achieve a near-perfect average score of 95.32, the state-of-the-art GPT-5 attains only 67.33. Crucially, we observe a "Scaling Law Paradox": simply increasing model parameters fails to yield linear improvements in these foundational visual capabilities. This study confirms that current MLLMs, despite their reasoning prowess, lack the essential physiological perceptual primitives required for generalized visual intelligence.

KidVis: Do Multimodal Large Language Models Possess the Visual Perceptual Capabilities of a 6-Year-Old?

TL;DR

KidVis systematically probes whether Multimodal LLMs possess foundational visual primitives by decomposing vision into six atomic capabilities and evaluating 20 MLLMs against a 6–7-year-old human baseline. The benchmark reveals a stark gap between human perceptual baselines and contemporary models, with proprietary systems lagging by roughly 28 points and open-source models performing often near chance on several tasks. A detailed analysis assigns failure modes to soft attention, architectural tokenization, semantic priors, and lack of grounded reference frames, while exposing a Scaling Law Paradox where increasing parameters fails to yield proportional perceptual gains. These findings suggest that achieving generalized visual intelligence requires architectural innovations and developmentally informed mechanisms beyond mere scale, emphasizing true physical perception grounding. The work provides a diagnostic framework and benchmarks to steer future research toward embodied-like visual perception and robust grounding in pixel-level reality.

Abstract

While Multimodal Large Language Models (MLLMs) have demonstrated impressive proficiency in high-level reasoning tasks, such as complex diagrammatic interpretation, it remains an open question whether they possess the fundamental visual primitives comparable to human intuition. To investigate this, we introduce KidVis, a novel benchmark grounded in the theory of human visual development. KidVis deconstructs visual intelligence into six atomic capabilities - Concentration, Tracking, Discrimination, Memory, Spatial, and Closure - already possessed by 6-7 year old children, comprising 10 categories of low-semantic-dependent visual tasks. Evaluating 20 state-of-the-art MLLMs against a human physiological baseline reveals a stark performance disparity. Results indicate that while human children achieve a near-perfect average score of 95.32, the state-of-the-art GPT-5 attains only 67.33. Crucially, we observe a "Scaling Law Paradox": simply increasing model parameters fails to yield linear improvements in these foundational visual capabilities. This study confirms that current MLLMs, despite their reasoning prowess, lack the essential physiological perceptual primitives required for generalized visual intelligence.
Paper Structure (29 sections, 4 figures, 2 tables)

This paper contains 29 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. KidVis Benchmark
  3. Classification of Visual Capabilities
  4. Visual Concentration (VC)
  5. Visual Tracking (VT)
  6. Visual Discrimination (VD)
  7. Visual Memory (VM)
  8. Visual Spatial (VS)
  9. Visual Closure (VCl)
  10. Task Instantiation
  11. Experiments
  12. Experimental Setup
  13. Multimodal Large Language Models
  14. Human Physiological Baseline
  15. Evaluation Metrics
  16. ...and 14 more sections

Figures (4)

  • Figure 1: The scheme of visual perceptual capabilities in KidVis benchmark.
  • Figure 2: Task Instantiations of KidVis Benchmark.
  • Figure 3: Comparison of representative MLLMs on KidVis benchmark. Left: Performance on 10 tasks within KidVis benchmark. Right: Performance on six visual perception capabilities.
  • Figure 4: Impact of open-source MLLMs parameters across three major families (Qwen3-VL, InternVL3.5, LLaVA-Next).