Video-SafetyBench: A Benchmark for Safety Evaluation of Video-Language Understanding

Video-SafetyBench is a benchmark dataset for evaluating the safety capabilities of video large vision-language models. It was created by BAAI for AI safety research and learning and was last updated on May 21, 2025. The dataset is licensed under CC BY-NC-SA 4.0.

Use Cases

• Benchmarking safety performance of video-language models based on the described safety evaluation tasks.
• Training models to identify unsafe content in video-language contexts based on the benchmark's examples.
• Conducting research on AI safety alignment for multimodal systems based on the dataset's focus.

Strengths

• Created by BAAI, a major AI research institute.
• Specifically designed for AI safety research in video-language understanding.
• License is explicitly stated as CC BY-NC-SA 4.0.

Limitations

• Dataset statistics and examples are referenced but not detailed in the provided input.
• Column-level documentation and sample data are unavailable, requiring inference after download.
• The data collection process is partially described but not fully detailed in the provided input.

Provenance

Source

BAAI

Collection Method

Data collection process is mentioned but not detailed in the provided input.

Freshness

Last updated 2025-05-21 02:28:49.