Dataset · MME-Benchmarks
Video-MME v2 Video Understanding Benchmark
MME-Benchmarks/Video-MME-v2
Updated benchmark with English videos and annotations for comprehensive evaluation of video models.
What's inside
The repository contains English annotations and video materials (from 1,000 to 10,000 examples) for evaluating video understanding. In the 2026 version, videos are transcoded to H265 without changing metrics, and task type markup has been added for logical coherence questions Q1–Q3.
What you can do with it
- Comprehensive testing of multimodal models for video content understanding
- Evaluating the logical coherence of VLM responses during video analysis
- Tasks
- Video text → text
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 13.3K
- Likes
- 52
How to load it
from datasets import load_dataset
ds = load_dataset("MME-Benchmarks/Video-MME-v2")