Video Dataset For Violence Detection In Multiple Contexts

Godseye Violence Detection Dataset is a composite resource assembled by valiantlynxz, merging three existing video datasets: RWF-2000, Hockey Violence Dataset, and Airtlab Violence Dataset. It contains videos depicting real-world fights, non-fight scenarios, violent hockey interactions, and violent incidents in various settings. The dataset was last updated on HuggingFace in November 2024.

Use Cases

• Train video classifiers to distinguish fight from non-fight scenarios using the labeled video frames from RWF-2000.
• Develop action recognition models for violent interactions in sports using the hockey-specific video clips.
• Benchmark multimodal models that fuse visual features from diverse settings like general surveillance and sports arenas.
• Fine-tune pre-trained networks on domain-specific violent incidents from the Airtlab dataset component.

Strengths

• Combines three established datasets, increasing scenario diversity
• Includes videos from both general real-world and specific sports contexts

Limitations

• Exact row count, video duration, and resolution are unknown
• Potential for class imbalance or label inconsistencies across the three merged sources
• Geographic and contextual coverage of the source datasets is unspecified

Provenance

Source

Composite of RWF-2000, Hockey Violence Dataset, and Airtlab Violence Dataset.

Collection Method

Aggregated from three existing public datasets.

Freshness

Last updated November 2024.