Holmes-VAD: Explainable Video Anomaly Detection via Multimodal LLMs

Holmes-VAD provides video sequences and textual reasoning labels for explainable anomaly detection, released by pipixin321 in 2025. It serves as the official data source for the Holmes-VAD framework, which integrates Multi-modal Large Language Models with surveillance footage. The dataset is distributed under the MIT license.

Use Cases

• Fine-tuning MLLMs to generate textual justifications for anomalous video segments
• Benchmarking explainable AI (XAI) performance using the provided reasoning labels
• Developing unbiased anomaly detection models that leverage multimodal context to reduce false positives

Strengths

• MIT license allows for flexible academic and commercial research
• Multimodal alignment between video frames and textual reasoning
• Official implementation of the Holmes-VAD paper

Limitations

• Metadata lacks explicit record counts or class distribution details
• Likely requires significant VRAM and compute resources to process the associated MLLM framework

Provenance

Source

pipixin321 GitHub repository

Collection Method

Expert-annotated video sequences for MLLM training

Freshness

Last updated March 2025.