IDD stands for Indian Driving Dataset, which is a road scene understanding dataset in an unstructured environment, mainly used for semantic segmentation and object detection in autonomous driving. The dataset contains 10,004 images from 182 drive sequences on Indian roads, which are divided into 34 categories and have fine annotations.

Compared with the mainstream dataset Cityscapes, this dataset has more annotations and a wider range of images, which is more conducive to solving new problems, such as domain adaptation, small sample learning, and behavior prediction in road scenes.