Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RSFAKE-1M: A Large-Scale Dataset for Detecting Diffusion-Generated Remote Sensing Forgeries

Zhihong Tan, Jiayi Wang, Huiying Shi, Binyuan Huang, Hongchen Wei, Zhenzhong Chen

TL;DR

RSFAKE-1M addresses the urgent need for robust detection of diffusion-generated remote sensing forgeries by creating a large-scale benchmark with 500k fake and 500k real RS images generated by ten diffusion models under six conditions. Real images are sourced from fMoW to match distribution, enabling controlled evaluation of generalization and robustness. Across extensive experiments, existing detectors struggle under domain shifts, but models trained on RSFAKE-1M show improved cross-dataset generalization and resilience to common degradations, highlighting the dataset’s utility for advancing RS forgery detection. This benchmark therefore provides a foundation for developing next-generation detectors tailored to diffusion-based RS forgeries and real-world deployment challenges.

Abstract

Detecting forged remote sensing images is becoming increasingly critical, as such imagery plays a vital role in environmental monitoring, urban planning, and national security. While diffusion models have emerged as the dominant paradigm for image generation, their impact on remote sensing forgery detection remains underexplored. Existing benchmarks primarily target GAN-based forgeries or focus on natural images, limiting progress in this critical domain. To address this gap, we introduce RSFAKE-1M, a large-scale dataset of 500K forged and 500K real remote sensing images. The fake images are generated by ten diffusion models fine-tuned on remote sensing data, covering six generation conditions such as text prompts, structural guidance, and inpainting. This paper presents the construction of RSFAKE-1M along with a comprehensive experimental evaluation using both existing detectors and unified baselines. The results reveal that diffusion-based remote sensing forgeries remain challenging for current methods, and that models trained on RSFAKE-1M exhibit notably improved generalization and robustness. Our findings underscore the importance of RSFAKE-1M as a foundation for developing and evaluating next-generation forgery detection approaches in the remote sensing domain. The dataset and other supplementary materials are available at https://huggingface.co/datasets/TZHSW/RSFAKE/.

RSFAKE-1M: A Large-Scale Dataset for Detecting Diffusion-Generated Remote Sensing Forgeries

TL;DR

RSFAKE-1M addresses the urgent need for robust detection of diffusion-generated remote sensing forgeries by creating a large-scale benchmark with 500k fake and 500k real RS images generated by ten diffusion models under six conditions. Real images are sourced from fMoW to match distribution, enabling controlled evaluation of generalization and robustness. Across extensive experiments, existing detectors struggle under domain shifts, but models trained on RSFAKE-1M show improved cross-dataset generalization and resilience to common degradations, highlighting the dataset’s utility for advancing RS forgery detection. This benchmark therefore provides a foundation for developing next-generation detectors tailored to diffusion-based RS forgeries and real-world deployment challenges.

Abstract

Detecting forged remote sensing images is becoming increasingly critical, as such imagery plays a vital role in environmental monitoring, urban planning, and national security. While diffusion models have emerged as the dominant paradigm for image generation, their impact on remote sensing forgery detection remains underexplored. Existing benchmarks primarily target GAN-based forgeries or focus on natural images, limiting progress in this critical domain. To address this gap, we introduce RSFAKE-1M, a large-scale dataset of 500K forged and 500K real remote sensing images. The fake images are generated by ten diffusion models fine-tuned on remote sensing data, covering six generation conditions such as text prompts, structural guidance, and inpainting. This paper presents the construction of RSFAKE-1M along with a comprehensive experimental evaluation using both existing detectors and unified baselines. The results reveal that diffusion-based remote sensing forgeries remain challenging for current methods, and that models trained on RSFAKE-1M exhibit notably improved generalization and robustness. Our findings underscore the importance of RSFAKE-1M as a foundation for developing and evaluating next-generation forgery detection approaches in the remote sensing domain. The dataset and other supplementary materials are available at https://huggingface.co/datasets/TZHSW/RSFAKE/.

Paper Structure

This paper contains 18 sections, 3 figures, 6 tables.

Table of Contents

  1. INTRODUCTION
  2. Related Work
  3. Remote Sensing Image Generation
  4. Remote Sensing Forgery Detection
  5. RSFAKE-1M Dataset
  6. Fake Image Generation
  7. Generation Model Selection.
  8. Image Generation.
  9. Real Image Collection
  10. Dataset Statistics
  11. Experiments
  12. Performance of Existing Detectors
  13. Cross-Dataset Generalization Analysis
  14. Intra-Dataset Generalization Analysis
  15. Cross Generator Evaluation.
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Overview of the RSFAKE-1M dataset generation pipeline. The dataset consists of 500,000 fake images generated by 10 diffusion-based models under six different conditions, and 500,000 real images sampled from the fMoW dataset.
  • Figure 2: Dataset composition of the collected synthetic remote sensing image corpus. From left to right: (a) train/val/test split, (b) proportion of fake and real samples, (c) distribution of fake generation conditions, (d) image size distribution, and (e) sample count per generative model.
  • Figure 3: Generalization and robustness evaluation of baseline models with ResNet-50 backbone trained on varying portions of RSFAKE-1M. The first two plots report accuracy on the RSFAKE-1M and DMAER test sets. The remaining three plots show robustness under common image degradations, including downsampling, JPEG compression, and Gaussian blur.