The Low Frequency Trap: Video Language Models Fail at Simple Event Bookkeeping
2026-08-06 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors created a new way to test how well video models count events by using simple, controlled videos with exact event logs. They found that models can count some types of events well when they happen slowly but struggle with quick or short events like blinks. Increasing video frame rates helps a bit but doesn’t fully fix the problem, and changing how the models are prompted doesn’t improve much either. Their method lets researchers pinpoint exactly where temporal understanding breaks down instead of just giving overall accuracy scores.
event countingvideo benchmarkstemporal reasoningparametric profilingexecutable event traceevent frequencyevent countmodel evaluationsampling rateprompting strategies
Authors
Sarvesh Baskar, Zikui Cai, Shayan Shabihi, Anirudh Satheesh, Muhammad R. Islam, Udari Madhushani Sehwag, Tom Goldstein, Furong Huang
Abstract
Real-world video benchmarks provide broad coverage, but their fixed clips entangle event count, rate, duration, and visual complexity, making failure modes hard to isolate. While existing programmatic benchmarks offer better control, they score only the final answer rather than auditing reported events against executable ground truth. To bridge this gap, we introduce trace-grounded parametric profiling for event counting in three controlled video tasks: bouncing-ball wall contacts, visual blinks, and categorical state transitions. Across 2,190 videos, we vary event count N and frequency F while holding rendering fixed. Each video includes an executable event trace for capability-surface estimation and timestamp-level evaluation. Our results reveal a staged temporal failure. At an 80% reliability threshold, Gemini 3.6 Flash reliably counts persistent state transitions up to 12 events at 0.5 and 1.0 Hz, yet demonstrates no reliable positive-count region for transient blinking events. Thus, event representation dictates whether a model initially accesses evidence -- a limitation that compounds as count and frequency increase. In the high-count, high-frequency regime, only 0.2% of final counts are correct and the model recovers just 18.1% of true events. To test if visual access is the primary bottleneck, we increase sampling rate. Although this boosts Bounce Ball accuracy from 19.6% to 29.3%, the reported sequence agrees with ground truth only 3.7% of the time. Extra frames can therefore inflate final scores without producing faithful event recovery. Different prompting strategies yield similarly limited gains, and real-world video evaluations show the same concentration of success at low event counts. Ultimately, trace-grounded profiling shifts video evaluation from aggregate accuracy metrics to a detailed diagnostic of where temporal reasoning fails.