Dataset · jiomimi
Anomaly detection and explanation benchmark for MLLM
jiomimi/UniAR-Bench
Over 40k Q&A pairs for evaluating multimodal models in video, image, and text anomaly detection.
What's inside
A cross-modal benchmark for evaluating multimodal language models (MLLM) built on top of open datasets. It contains over 40,199 question-answer pairs covering video, image, text, and tabular data, grouped into four general tasks for anomaly detection and detailed reasoning.
What you can do with it
- Evaluating MLLM capability to locate and reason about anomalies in video and images
- Comparing multimodal models in cross-modal reasoning tasks
- Size
- 10K–100K rows
- Downloads this month
- 34.2K
- Likes
- 0
- Access
- on request on Hugging Face
How to load it
from datasets import load_dataset
ds = load_dataset("jiomimi/UniAR-Bench")