LAV-DF: Localized Audio-Visual DeepFake Benchmark

LAV-DF is a dataset for content-driven audio-visual deepfake detection and temporal forgery localization. The dataset was created by ControlNet for research presented at DICTA and submitted to CVIU. It was last updated on Hugging Face on October 16, 2024.

Use Cases

• Training models for temporal forgery localization based on synchronized audio-visual content.
• Benchmarking content-driven audio-visual deepfake detection methods.
• Developing multimodal methods to identify manipulated segments within videos.

Strengths

• Dataset is associated with a paper that received a Best Award at DICTA.
• Dataset is designed for a specific, advanced task of temporal forgery localization.
• The dataset page was updated on 2024-10-16, indicating recent maintenance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license are unknown, which may limit suitability assessment.

Provenance

Source

ControlNet

Freshness

Last updated 2024-10-16 11:12:33