CinematicVQA: Benchmarking Film-Grammar Reasoning in Large Vision-Language Models

2026-09-23

AI Summary

This study addresses the limitation of existing video question-answering benchmarks, which predominantly focus on low-level visual recognition while overlooking the narrative influence of cinematic grammar. To this end, we propose the first evaluation benchmark specifically designed for cinematic grammar reasoning, introducing structured cinematic scene graphs to assess the narrative reasoning capabilities of vision-language models (VLMs). Through systematic experiments combining chain-of-thought prompting with supervised fine-tuning, our findings reveal that current VLMs perform better on descriptive tasks than on recognition tasks, yet chain-of-thought prompting fails to effectively bridge the semantic gap. In contrast, supervised fine-tuning substantially enhances both narrative comprehension and multi-hop reasoning performance. This work fills a critical gap in film-level narrative reasoning evaluation, exposing the semantic bottlenecks of VLMs and highlighting the pivotal role of fine-tuning in advancing their narrative understanding.

Abstract

Cinematography, the craft of visual storytelling through framing, lighting, and camera operation, fundamentally shapes how audiences perceive and emotionally engage with video content. While Large Vision Language Models (LVLMs) have made remarkable progress in video question answering, existing benchmarks primarily focus on identifying low-level techniques rather than understanding their storytelling impact. To address this, we introduce CinematicVQA, the first-of-its-kind benchmark for cinematic video understanding that goes beyond technique recognition to evaluate film-grammar reasoning, utilizing our introduced Cinematic Scene Graph (CSG), a structured representation that links filming techniques to their perceptual effects and narrative functions. Through comprehensive evaluation of state-of-the-art LVLMs, we reveal a striking semantic gap: models consistently perform higher on describing visual presentations than on identifying the underlying techniques. Surprisingly, Chain-of-Thought prompting fails to provide consistent gains and degrades performance for most models, suggesting that current LVLMs lack sufficient cinematic domain knowledge to benefit from step-by-step reasoning. Fine-tuning on \textsc{CinematicVQA-train} yields consistent improvements, particularly for narrative function and multi-hop reasoning. Overall, \textsc{CinematicVQA} serves both as a rigorous benchmark for cinematic evaluation in LVLMs and as a practical dataset for training more film-aware video models.

Problem

• Large Vision-Language Models
• Video Question Answering
• Film-Grammar Reasoning
• Cinematic Understanding
• Semantic Gap

Innovation

• CinematicVQA
• Cinematic Scene Graph
• Film-Grammar Reasoning
• Large Vision-Language Models
• Video Question Answering