media-deepfake-benchmark

What problem does it solve?

Automates generation and validation of deepfake benchmark samples across five manipulation categories to ensure comprehensive coverage and operational detail.

Core Features & Use Cases

• Automated benchmark task: model fills three fields for five categories across media manipulation, enabling end-to-end validation of coverage.
• Validation via pytest-like checks: ensures minimum word counts, placeholder removal, and marker-based specificity for each sample.
• Standardized format: anchors outputs to a consistent schema suitable for evaluation pipelines in media forensics research.

Quick Start

Run the benchmark to generate five detailed samples and validate coverage with a ViT-based classifier.