Deepfake Unfiltered 30K: Binary Quality Labels for Deepfake Images and Videos

Deepfake-Unfiltered-30K is a dataset for training models to assess the quality of deepfake images and videos. It contains binary labels where 0 indicates a bad-quality deepfake and 1 indicates a good-quality deepfake. The dataset was created by prithivMLmods and was last updated on 2025-04-19.

Use Cases

• Train binary classifiers for deepfake quality assessment based on the provided good/bad labels.
• Improve deepfake detection systems by filtering or prioritizing analysis based on quality scores.
• Enhance deepfake generation techniques by learning from quality-labeled examples.
• Benchmark model performance on the task of automated deepfake quality evaluation.

Strengths

• Provides a clear binary classification task (good/bad quality) for model training.
• Dataset was updated on 2025-04-19, suggesting recent maintenance.
• Explicitly designed for deepfake quality assessment, a specific computer vision task.

Limitations

• Row count, file formats, and column-level documentation are unknown, limiting suitability assessment.
• Description metadata is limited; actual data quality and sample composition require manual inspection after download.

Provenance

Source

huggingface user prithivMLmods

Collection Method

Likely collected or generated for training deepfake quality assessment models.

Time Range

null

Freshness

Last updated 2025-04-19 09:32:23

Geography

null