Building Detection Dataset-Satellite Imagery

This dataset contains building scenes in satellite imagery captured at multiple spatial resolutions (ranging from 0.3 m to 1 cm per pixel) and includes diverse lighting conditions and seasonal variations across both urban and rural environments.The images are derived from large format tiles obtained from OpenAerialMap and Google Earth.

The dataset contains a total of 3730 images for training and validation. The annotations are in YOLOv8 format.

The training dataset contains a total of 2984, and the validation dataset contains 746 images. 

test-1 is the unseen data containing 15 images of complex building scenes.

test-2 is an extended version of test-1 containing 20 images for extended evaluation.

The Complete_data file contains the full dataset without any predefined splits. Researchers can use it to create their own independent training, validation, and test subsets according to their experimental requirements.