Reproduce “RSFAKE-1M: A Large-Scale Dataset for Detecting Diffusion-Generated Remote Sensing Forgeries” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhihong Tan, Jiayi Wang, Huiying Shi, Binyuan Huang, Hongchen Wei, Zhenzhong Chen
Researchers from Wuhan University developed RSFAKE-1M, a comprehensive dataset of one million images, specifically to improve the detection of sophisticated remote sensing forgeries created by diffusion models. This dataset, comprising 500,000 synthetic and 500,000 authentic images, facilitates the development of more robust and generalizable forgery detection systems.
Key insights
- Existing state-of-the-art forgery detectors, both general and remote sensing-specific, show considerable performance degradation on diffusion-generated remote sensing forgeries.
- Training models on RSFAKE-1M consistently leads to superior cross-dataset generalization compared to training on existing smaller datasets.
- Detection performance is influenced by the specific generative model and the generation condition (e.g., text-only, image-guided), indicating distinct forgery patterns.
- Increasing the scale of training data, as provided by RSFAKE-1M, directly correlates with improved in-domain and cross-domain detection performance and enhanced robustness against image degradations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.