Overview
The Cityscapes Dataset is a comprehensive resource designed for advancing research in semantic urban scene understanding. It features a diverse collection of stereo video sequences recorded in street scenes across 50 different cities. The dataset provides high-quality, pixel-level annotations for 5,000 frames, complemented by a larger set of 20,000 weakly annotated frames. Cityscapes aims to facilitate the development and evaluation of vision algorithms for tasks such as pixel-level, instance-level, and panoptic semantic labeling. It supports research focused on leveraging large volumes of annotated data, particularly for training deep neural networks, offering rich metadata including preceding and trailing video frames, stereo information, GPS data, and vehicle odometry. The dataset is freely available for academic and non-commercial purposes.
Common tasks