DOTA is a large-scale dataset for object detection in aerial images. This dataset is version v1.5 and v2.0.

The dataset contains aerial images from different sensors and platforms. The pixel size of each image ranges from 800 × 800 to 20,000 × 20,000, and contains objects of different sizes, orientations, and shapes. The instances in the DOTA images are labeled by an aerial image interpretation expert using arbitrary quadrilaterals (8 degrees of freedom).

There are three versions of this dataset:

• DOTA-v1.0 It contains 15 common categories, 2,806 images and 188,282 instances. The ratios of training set, validation set and test set are 1/2, 1/6 and 1/3 respectively.
• DOTA-v1.5 contains the same images, but with annotations added for very small instances (less than 10 pixels). In addition, a new class “container crane” has been added. A total of 403,318 instances are included.
• DOTA-v2.0 has a total of 18 common categories, 11,268 images (divided into training, validation, test-dev, and test-challenge sets), and 1,793,658 instances.