DOTA (Dataset of Object deTection in Aerial images) is a large-scale dataset for object detection in aerial images that can be used to develop and evaluate object detectors from high-altitude cameras. The images are collected from different sensors and platforms. Each image is of the size in the range from 800 × 800 to 20,000 × 20,000 pixels and contains objects exhibiting a wide variety of scales, orientations, and shapes. The instances in DOTA images are annotated by experts in aerial image interpretation by arbitrary (8 d.o.f.) quadrilateral.
The dataset contains 15 common categories (e.g., ship, plane, vehicle, swimming pool, etc.), which are annotated on the images as (rotated) bounding boxes defined by four pairs of points. The data is split into the following subsets:
- Training data: 1411 images (~1/2 of the available data)
- Validation data: 458 images (~1/6 of the available data)
- Test data: (~1/6 of the available data)
The annotated objects have a minimum size of 10 pixels.
Annotation type: object detection
Created by: Xia, et. al., https://arxiv.org/pdf/2102.12219.pdf
Date published: 2018
License: Google Earth, Inc.
Dataset size: 2,806 images and 188, 282 instances