JamalLee/Omni-Fake-OOD

Omni-Fake-OOD

Omni-Fake-OOD is the out-of-distribution benchmark split of Omni-Fake. Samples come from held-out generators and platforms not included in training, for measuring cross-domain generalization. It covers image, audio, video, and audio–video talking-head (AV-TH) with the same three-class labels as Omni-Fake-SET: real, fully synthetic, and tampered. Use together with Omni-Fake-SET (in-distribution training data).

• —Paper: arXiv:2605.01638
• —Project page: Omni-Fake
• —License: CC-BY-4.0

Video (hybrid release)

OOD Video totals 22,000 clips: 1,000 real + 1,000 full_synthetic + 20,000 tampered.

Quick start

from datasets import load_dataset

# Image (default config)
ds_img = load_dataset("JamalLee/Omni-Fake-OOD", "image")
print(ds_img)

# Audio
ds_aud = load_dataset("JamalLee/Omni-Fake-OOD", "audio")

# Video tampered (parquet)
ds_vid = load_dataset("JamalLee/Omni-Fake-OOD", "video")
# Real + full_synthetic: extract data/Video/video-ood.7z

# Audio–video talking head
ds_avth = load_dataset("JamalLee/Omni-Fake-OOD", "avth")

Fields (by modality)

Image: image, mask, label, generator, filename, split

Audio: audio, label, generator, filename, split, spoof_intervals

Video (parquet, tampered only): video, label, generator, filename, split

Video (`.7z`, real + full_synthetic): extract video-ood.7z; benchmark layout → real / full_synthetic

AV-TH: video, label, generator, filename, split

Labels are real, full_synthetic, or tampered (same three-class scheme for Image and Video).

Dataset Viewer

Open the Dataset Viewer tab and keep subset `image` selected (default). The image column renders as thumbnails; mask shows tamper regions when available. Switch subset to video, audio, or avth for other modalities.

Citation

@article{li2026omnifake,
  title={Omni-Fake: Benchmarking Unified Multimodal Social Media Deepfake Detection},
  author={Li, Tianxiao and Huang, Zhenglin and Wen, Haiquan and others},
  journal={arXiv preprint arXiv:2605.01638},
  year={2026}
}