Dataset · ControlNet
Audio-visual deepfake dataset with timestamps
ControlNet/LAV-DF
A benchmark of English videos for detecting and localizing precise moments of audio and video tampering.
What's inside
Includes between 100K and 1 million English video recordings with content-driven deepfakes. The dataset was prepared for academic research presented at the DICTA conference and CVIU journal. Its main feature is time-interval annotation, enabling the detection not only of deepfake presence but also the specific seconds of manipulation.
What you can do with it
- Training audio-visual deepfake detectors
- Training models for temporal localization of tampering in video streams
- Benchmarking multimodal content safety systems
- Tasks
- Video classification
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 1.3K
- Likes
- 22
- Access
- on request on Hugging Face
How to load it
from datasets import load_dataset
ds = load_dataset("ControlNet/LAV-DF")