Investigating Video Reasoning Capability of Large Language Models with Tropes in Movies

TL;DR

This paper introduces the Tropes in Movies dataset to evaluate large language models' video reasoning skills, revealing current methods' limitations and proposing enhancements that improve reasoning performance but still fall short of human levels.

Contribution

The paper presents a new dataset and evaluation protocol for video reasoning in LLMs, along with novel methods FEVoRI and ConQueR that improve reasoning accuracy.

Findings

Current models only marginally outperform random baselines.

Proposed methods improve F1 scores by 15 points.

Performance remains below human levels (40 vs. 65 F1).

Abstract

Large Language Models (LLMs) have demonstrated effectiveness not only in language tasks but also in video reasoning. This paper introduces a novel dataset, Tropes in Movies (TiM), designed as a testbed for exploring two critical yet previously overlooked video reasoning skills: (1) Abstract Perception: understanding and tokenizing abstract concepts in videos, and (2) Long-range Compositional Reasoning: planning and integrating intermediate reasoning steps for understanding long-range videos with numerous frames. Utilizing tropes from movie storytelling, TiM evaluates the reasoning capabilities of state-of-the-art LLM-based approaches. Our experiments show that current methods, including Captioner-Reasoner, Large Multimodal Model Instruction Fine-tuning, and Visual Programming, only marginally outperform a random baseline when tackling the challenges of Abstract Perception and Long-range Compositional Reasoning. To address these deficiencies, we propose Face-Enhanced Viper of Role Interactions (FEVoRI) and Context Query Reduction (ConQueR), which enhance Visual Programming by fostering role interaction awareness and progressively refining movie contexts and trope queries during reasoning processes, significantly improving performance by 15 F1 points. However, this performance still lags behind human levels (40 vs. 65 F1). Additionally, we introduce a new protocol to evaluate the necessity of Abstract Perception and Long-range Compositional Reasoning for task resolution. This is done by analyzing the code generated through Visual Programming using an Abstract Syntax Tree (AST), thereby confirming the increased complexity of TiM. The dataset and code are available at: https://ander1119.github.io/TiM