CHAI: WWII Crater Detection in Historical Aerial Images

99 historical aerial images from Austria and Germany, captured between 1943 and 1945, form the core of this dataset. It provides annotations for impact craters and includes three versions: raw originals, and two derived datasets of 960×960 patches formatted for machine learning evaluation. The data is intended as a benchmark for computer vision tasks like object detection in historical geospatial imagery.

Use Cases

• Training object detection models for crater identification based on annotated bounding boxes.
• Benchmarking model performance on historical, non-standardized aerial imagery.
• Studying geospatial patterns of bomb impacts during WWII in Central Europe.
• Developing preprocessing techniques for historical image normalization and patch extraction.

Strengths

• Provides manually refined crater annotations, suggesting high-quality labels for model training.
• Includes multiple dataset versions (raw, full, light) tailored for different research and evaluation needs.
• Covers a specific, historically significant time range (1943-1945) from two European countries.

Limitations

• The dataset is small in scale, with only 99 original source images.
• Key metadata such as exact row counts, file sizes, and last update dates are not provided across sources.
• Access is restricted to non-commercial research and requires a signed usage agreement.

Provenance

Source

Marvin Burges, TU Wien

Collection Method

Aerial imagery collection from 1943-1945, followed by manual crater annotation and dataset derivation for a WACV2024 publication.

Time Range

1943 - 1945

Geography

Austria and Germany