STER-VLM: Spatio-Temporal With Enhanced Reference Vision-Language Models

TL;DR

STERVLM is a resource-efficient vision-language framework that improves fine-grained spatio-temporal understanding for traffic analysis by combining caption decomposition, frame selection, reference-driven understanding, and prompt techniques.

Contribution

It introduces a novel, computationally efficient approach that enhances VLMs for traffic scene understanding through multiple innovative techniques.

Findings

Achieved a test score of 55.655 in AI City Challenge 2025 Track 2.

Demonstrated substantial improvements in semantic richness and scene interpretation.

Validated effectiveness on WTS and BDD datasets.

Abstract

Vision-language models (VLMs) have emerged as powerful tools for enabling automated traffic analysis; however, current approaches often demand substantial computational resources and struggle with fine-grained spatio-temporal understanding. This paper introduces STER-VLM, a computationally efficient framework that enhances VLM performance through (1) caption decomposition to tackle spatial and temporal information separately, (2) temporal frame selection with best-view filtering for sufficient temporal information, and (3) reference-driven understanding for capturing fine-grained motion and dynamic context and (4) curated visual/textual prompt techniques. Experimental results on the WTS \cite{kong2024wts} and BDD \cite{BDD} datasets demonstrate substantial gains in semantic richness and traffic scene interpretation. Our framework is validated through a decent test score of 55.655 in the AI City Challenge 2025 Track 2, showing its effectiveness in advancing resource-efficient and accurate traffic analysis for real-world applications.