Synthetic and Manipulated Overhead Imagery For Forensics Research

Brandon B. May, Kirill Trapeznikov, Shengbang Fang, Matthew C. Stamm
Comprehensive Dataset of Synthetic and Manipulated Overhead Imagery For Development and Evaluation of Forensic Tools
https://arxiv.org/abs/2305.05784

@inproceedings{may2023comprehensive,
  title={Comprehensive Dataset of Synthetic and Manipulated Overhead Imagery for Development and Evaluation of Forensic Tools},
  author={May, Brandon B and Trapeznikov, Kirill and Fang, Shengbang and Stamm, Matthew},
  booktitle={Proceedings of the 2023 ACM Workshop on Information Hiding and Multimedia Security},
  pages={145--150},
  year={2023}
}

👉 ⬇️ Dataset Download

• Dataset Overview
  • Sources
  • Inpainting Category
  • Inpainting Size
  • Fully Synthetic Categories
  • Dataset Stats
  • Dataset Download
  • Partial Manipulations
  • Natural Disaster Simulation
• Contact
• Acknowledgment

Dataset Overview

Sources

Pristine Tile Providers

City Conditioning

Inpainting Category

Inpainting Category	Region Type	Manipulated 🔄 Pristine	Mask
Greenspace	Bezier
Buildings	Bezier
Greenspace	GraphCut
Buildings	GraphCut

Inpainting Size

Inpainting Area Size	Manipulated 🔄 Pristine	Mask
Exra Small
Small
Medium
Large

Fully Synthetic Categories

Dataset Stats

Train Set

The train set consists of 10,000 images in total:

• 4,964 pristine (unmanipulated)
• 2,539 fully synthetic
• 2,497 partially manipulated (with manipulated regions between 1/16 and 1/4 of the image area)
  • 1,260 buildings_roads manipulation class
  • 1,237 greenspace_water manipulation class

Each image has a corresponding binary mask image that indicates the manipulated region with the color white.

Test Set

The test set consists of 3,150 images in total:

• 1,511 pristine (unmanipulated)
• 751 fully synthetic
• 888 partially manipulated (including 150 with manipulated regions < 1/16 of the image area)
  • 432 buildings_roads manipulation class
  • 456 greenspace_water manipulation class

Each image has a corresponding binary mask image that indicates the manipulated region with the color white.

Dataset Download

⬇️ Dataset Download

Metadata CSV Format

Each dataset is provided with a metadata.csv file which contains detailed information about each image. The following columns are particularly relevant for training and testing:

• image_filename: Relative path to the image, i.e. im0000.png
• mask_filename: Relative path to the corresponding binary mask, i.e. im0000_mask.png
• manipulated_fraction: [0-1], Fraction of the image that has been manipulated
• manipulation_class:
  • buildings_roads
  • greenspace_water
  • Empty for pristine and fully synthetic images
• manipulation_type:
  • pristine: Unmanipulated
  • generated_uncond: Fully synthetic, no basemap used
  • generated_cond: Fully synthetic, generated basemap used
  • generated_real: Fully synthetic, real basemap used
  • inpainted_struct: Partially manipulated, buildings_roads
  • inpainted_unstruct: Partially manipulated, greenspace_water

Additional Generation Modes

Partial Manipulations

Natural Disaster Simulation

Contact

For more information: Kirill Trapeznikov, kirill.trapeznikov@str.us; Brandon May, brandonbmay@gmail.com

Acknowledgment

This material is based upon work supported by DARPA under Contract No. HR0011-20-C-0129. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of DARPA.

DISTRIBUTION A. Approved for public release: distribution unlimited.