Dataset · Jinxing1
MQ-RAVSBench: mask audit benchmark
Jinxing1/MQ-RAVSBench
An audio-video benchmark with masks for evaluating referring audiovisual segmentation quality.
What's inside
A benchmark ranging from 1K to 10K examples in English, containing video clips, audio, referring expressions, ground truth object masks, and candidate masks with various types of errors. Used for auditing and validating segmentation model solutions (e.g., MQ-Auditor).
What you can do with it
- Benchmarking and evaluating audiovisual segmentation models
- Testing mask filtering and correction algorithms
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 27.9K
- Likes
- 2
How to load it
from datasets import load_dataset
ds = load_dataset("Jinxing1/MQ-RAVSBench")