PerceptionTest

The 'PerceptionTest' is a benchmark used to evaluate video question-answering performance, specifically assessing the ability of models to understand and interpret visual information in videos.

Papers using PerceptionTest (4)

Pushupbench: Your VLM Is Not Good At Counting Pushups2026V-JEPA 2: Self-supervised Video Models Enable Understanding, Prediction And Planning2025 · 5 citesBoxTuning: Directly Injecting the Object Box for Multimodal Model Fine-Tuning2026MACD: Model-Aware Contrastive Decoding via Counterfactual Data2026