VGenST-Bench: A Benchmark for Spatio-Temporal Reasoning via Active Video Synthesis

TL;DR

VGenST-Bench is a new benchmark for evaluating spatio-temporal reasoning in multimodal models, created through active video synthesis to provide diverse and controlled scenarios.

Contribution

It introduces a generative, multi-agent pipeline for creating a comprehensive, high-quality video benchmark with a hierarchical task suite for detailed reasoning assessment.

Findings

Enables fine-grained diagnosis of spatio-temporal understanding.

Provides diverse scenarios through a 3x2x2 taxonomy.

Ensures quality with human-in-the-loop synthesis.

Abstract

Spatio-temporal reasoning is a core capability for Multimodal Large Language Models (MLLMs) operating in the real world. As such, evaluating it precisely has become an essential challenge. However, existing spatio-temporal reasoning benchmark datasets primarily rely on static image sets or passively curated video data, which limits the evaluation of fine-grained reasoning capabilities. In this paper, we introduce VGenST-Bench, a video benchmark that employs generative models to actively synthesize highly controlled and diverse evaluation scenarios. To construct VGenST-Bench, we propose a multi-agent pipeline incorporating a human quality control stage, ensuring the quality of all generated videos and QA pairs. We establish a comprehensive 3x2x2 video taxonomy, encompassing Spatial Scale, Perspective, and Scene Dynamics to span diverse scenarios. Furthermore, we design a hierarchical task suite that decouples low-level visual perception from high-level spatio-temporal reasoning. By shifting the paradigm from passive curation to active synthesis, VGenST-Bench enables fine-grained diagnosis of spatio-temporal understanding in MLLMs.