10,000 images of faces labeled as real or fake, intended for training AI classifiers. The fake images are generated using NVIDIA's StyleGAN3 model. The dataset's provenance, size, and license details are unknown.
Use Cases
- Train a binary classifier based on the 'real vs fake' labels.
- Benchmark generative model detection algorithms based on StyleGAN3-generated images.
- Develop adversarial training techniques based on synthetic face data.
- Evaluate model robustness against state-of-the-art generative AI outputs.
Strengths
- Contains 10,000 labeled images.
- Focuses on a specific, modern generative model (StyleGAN3).
Limitations
- Row count and file size are unknown, which may limit suitability assessment.
- Column-level documentation is absent; field semantics must be inferred after download.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Collection Method
- Fake images are generated using NVIDIA's StyleGAN3 model.