Time Blindness: Why Video-Language Models Can't See What Humans Can?

TL;DR

Video-language models excel at spatial understanding but struggle with purely temporal patterns, revealing a significant gap compared to human perception, especially in noise-like sequences.

Contribution

We introduce SpookyBench, a benchmark highlighting VLMs' inability to interpret temporal sequences without spatial cues, and analyze the limitations across models.

Findings

Humans recognize patterns in noise-like sequences with over 98% accuracy.

State-of-the-art VLMs achieve 0% accuracy on SpookyBench.

Temporal understanding in VLMs degrades faster than in humans under low spatial SNR.

Abstract

Recent advances in vision-language models (VLMs) have made impressive strides in understanding spatio-temporal relationships in videos. However, when spatial information is obscured, these models struggle to capture purely temporal patterns. We introduce $\textbf{SpookyBench}$, a benchmark where information is encoded solely in temporal sequences of noise-like frames, mirroring natural phenomena from biological signaling to covert communication. Interestingly, while humans can recognize shapes, text, and patterns in these sequences with over 98% accuracy, state-of-the-art VLMs achieve 0% accuracy. This performance gap highlights a critical limitation: an over-reliance on frame-level spatial features and an inability to extract meaning from temporal cues. Furthermore, when trained in data sets with low spatial signal-to-noise ratios (SNR), temporal understanding of models degrades more rapidly than human perception, especially in tasks requiring fine-grained temporal reasoning. Overcoming this limitation will require novel architectures or training paradigms that decouple spatial dependencies from temporal processing. Our systematic analysis shows that this issue persists across model scales and architectures. We release SpookyBench to catalyze research in temporal pattern recognition and bridge the gap between human and machine video understanding. Dataset and code has been made available on our project website: https://timeblindness.github.io/.