Can Vision-Language Models Solve the Shell Game?

TL;DR

This paper introduces VET-Bench, a synthetic testbed revealing current vision-language models' inability to track identical objects over time, and proposes SGCoT to significantly improve their tracking performance.

Contribution

The paper presents VET-Bench for diagnosing tracking limitations in VLMs, provides a theoretical analysis of model constraints, and introduces SGCoT to enhance tracking accuracy to over 90%.

Findings

Current VLMs perform at chance level on VET-Bench.

Fixed-depth transformers are fundamentally limited in tracking indistinguishable objects.

SGCoT achieves over 90% accuracy on VET-Bench.

Abstract

Visual entity tracking is an innate cognitive ability in humans, yet it remains a critical bottleneck for Vision-Language Models (VLMs). This deficit is often obscured in existing video benchmarks by visual shortcuts. We introduce VET-Bench, a synthetic diagnostic testbed featuring visually identical objects that necessitate tracking exclusively through spatiotemporal continuity. Our experiments reveal that current state-of-the-art VLMs perform at or near chance level on VET-Bench, exposing a fundamental limitation: an over-reliance on static frame-level features and a failure to maintain entity representations over time. We provide a theoretical analysis drawing connections to the state-tracking problem, proving that fixed-depth transformer-based VLMs are fundamentally limited in tracking indistinguishable objects without intermediate supervision due to expressivity constraints. To address this, we propose Spatiotemporal Grounded Chain-of-Thought (SGCoT): generating object trajectories as explicit intermediate states. Leveraging Molmo2's object tracking ability, we elicit SGCoT reasoning by fine-tuning on synthesized text-only data for alignment. Our method achieves state-of-the-art accuracy exceeding 90% on VET-Bench, demonstrating that VLMs can reliably solve the video shell-game task end-to-end without external tools. Our code and data are available at https://vetbench.github.io .