VA-SafetyBench: Video and Audio Safety Benchmark Dataset

VA-SafetyBench is a dataset of video and audio files compiled for safety benchmarking. The dataset was authored by luweikai and was last updated on June 2, 2025. It is hosted on the Hugging Face platform and includes a warning about potentially sensitive or harmful content.

Use Cases

• Benchmarking multimodal AI model safety based on video content.
• Evaluating audio processing systems for harmful or sensitive content.
• Training safety classifiers for video inputs.
• Developing red-teaming protocols for video and audio AI systems.

Strengths

• Last updated on June 2, 2025, indicating recent maintenance.
• Explicitly warns users about sensitive content, promoting responsible use.

Limitations

• Dataset size, row count, and file formats are unknown, limiting suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

luweikai

Freshness

Last updated 2025-06-02 22:43:09