scu-zjz/IMDLBenCo

What it solves

IMDL-BenCo provides a standardized framework for Image Manipulation Detection and Localization (IMDL). It addresses the lack of a comprehensive, modular codebase and benchmark for identifying whether an image has been tampered with and precisely locating the manipulated regions at a pixel level.

How it works

The project decomposes the IMDL pipeline into standardized, reusable components. This modular approach allows researchers to easily customize and build model construction pipelines. The codebase incorporates training code for multiple state-of-the-art (SOTA) models, establishing a comprehensive benchmark for comparing different forensic techniques.

Who it’s for

Computer vision researchers and forensic analysts who are working on detecting and localizing image forgeries, including AI-generated content and traditional image manipulation.

Highlights

• Modular Architecture: Decomposes the framework into reusable components to improve coding efficiency and customization.
• Comprehensive Benchmark: Includes implementations of various SOTA models to provide a reliable baseline for the field.
• PyPI Distribution: Available as a Python library (pip install imdlbenco) for easy installation and integration.
• Academic Recognition: Accepted as a Spotlight paper at NeurIPS 2024 Track Datasets and Benchmarks.