MVBench

MVBench is a benchmark dataset used to evaluate the performance of vision-language models (VLMs) in tasks related to video understanding and temporal reasoning, specifically focusing on counting repetitions in video clips.

Papers using MVBench (13)

Pushupbench: Your VLM Is Not Good At Counting Pushups2026VisReflect: Latent Visual Reflection for Fine-Grained Perception in Long Visual Context2026LongVPO: From Anchored Cues to Self-Reasoning for Long-Form Video Preference Optimization2026Not All Modalities Are Equal: Instruction-Aware Gating for Multimodal Videos2026Clue Matters: Leveraging Latent Visual Clues to Empower Video Reasoning2026MACD: Model-Aware Contrastive Decoding via Counterfactual Data2026MSVBench: Towards Human-Level Evaluation of Multi-Shot Video Generation2026Improving Video Question Answering through query-based frame selection2026Video Evidence to Reasoning Efficient Video Understanding via Explicit Evidence Grounding2026Enhancing Temporal Understanding In Video-llms Through Stacked Temporal Attention In Vision Encoders2025Ro-bench: Large-scale Robustness Evaluation Of Mllms With Text-driven Counterfactual Videos2025Seeing Is Not Reasoning: Mvpbench For Graph-based Evaluation Of Multi-path Visual Physical Cot2025Gam-agent: Game-theoretic And Uncertainty-aware Collaboration For Complex Visual Reasoning2025