DOTA v2.0

DOTA-v2.0 收集了更多的谷歌地球、GF-2 卫星和航拍图像。 DOTA-v2.0 共有 18 个常用类别、11,268 个图像和 1,793,658 个实例。相比DOTA-v1.5，它进一步增加了“机场”和“直升机停机坪”的新类别。 DOTA 的 11,268 张图像分为训练集、验证集、测试开发集和测试挑战集。为了避免过拟合的问题，训练集和验证集的比例要小于测试集。此外，我们有两个测试集，即 test-dev 和 test-challenge。训练包含 1,830 张图像和 268,627 个实例。验证包含 593 个图像和 81,048 个实例。我们发布了用于训练和验证集的图像和基本事实。 Test-dev 包含 2,792 个图像和 353,346 个实例。我们发布了图像，但没有发布基本事实。测试挑战包含 6,053 个图像和 1,090,637 个实例。测试挑战的图像和基本事实仅在挑战期间可用。

DOTA-v2.0 incorporates more Google Earth, GF-2 satellite, and aerial imagery. DOTA-v2.0 consists of 18 common categories, 11,268 images, and 1,793,658 instances. Compared with DOTA-v1.5, it additionally introduces two new categories: "airport" and "helipad". The 11,268 images in DOTA are divided into training set, validation set, test-dev set, and test-challenge set. To avoid overfitting, the combined size of the training and validation sets is smaller than that of the test sets. In addition, there are two test sets: test-dev and test-challenge. The training set contains 1,830 images and 268,627 instances. The validation set contains 593 images and 81,048 instances. We have released the images and ground truths for the training and validation sets. The test-dev set contains 2,792 images and 353,346 instances. We have released the images but not the ground truths for the test-dev set. The test-challenge set contains 6,053 images and 1,090,637 instances. The images and ground truths for the test-challenge set are only available during the challenge period.