ForensicHub Documentation
Unified Benchmark and Codebase for Fake Image Detection and Localization
Full Task Compatibility
Supports four major image forensic tasks including Deepfake, IMDL, AIGC, and Document, covering image-level detection and pixel-level localization, truly achieving task unification and domain integration.
Free Combination Pipeline
Decouples datasets, preprocessors, models, and evaluators through a modular architecture, allowing users to freely combine and build cross-task training and evaluation processes based on YAML configurations.
Standardization and Scalability
Provides unified interfaces and protocols, adapts to existing benchmarks (such as DeepfakeBench and IMDLBenCo), and supports the extension of custom datasets, models, and metrics, balancing engineering reproduction and scientific exploration.
For more details, please refer to the Guide.