Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
video
video
3.08
6.12
label
class label
2 classes
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
1vocoded
1vocoded
1vocoded
1vocoded
1vocoded
1vocoded
1vocoded
1vocoded
1vocoded
1vocoded
1vocoded
1vocoded
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
0main
1vocoded
1vocoded
1vocoded
1vocoded
End of preview. Expand in Data Studio

MOZAIC review sample (private)

A seeded sample of the MOZAIC release for the last human review. The review page (blind test, then pair review) is the Space Deep-Fake/mozaic-review; a simple one-page viewer of these clips is the Space Deep-Fake/mozaic-sample-viewer. Not a release and not for sharing: VoxCeleb2-derived clips of real people, including generated fakes of them.

  • 60 pairs, 360 clips, byte-identical to the release (sha256 in clips.jsonl): 50 drawn at random (10 per generator and length; per gender 3 train, 1 val, 1 test) and 10 edge pairs (per generator and length, the pair closest to a face or picture limit and the pair closest to a voice limit).
  • review.jsonl: one row per pair: why it was drawn, its six clips and the target person's two reference clips (VoxCeleb2 ids).
  • clips.jsonl: the release's rows for these clips (all fields, with scores).
  • The 40 blind-test clips are on the Space only, under neutral names; their answer key is not on Hugging Face.
  • Drawn 2026-10-04 12:05:25, seed 20261004, from the release whose checksum list has sha256 1a4e7a9a65580b52e7204b0fe2adc416bb75a8f08b3fdaaa83e9fa91fc1f51c1.
Downloads last month
2