The Cityscapes dataset is one of the most large-scale datasets of stereo videos featuring urban scenes. It contains recordings from 50 different cities of Germany with high-quality pixel-accurate annotations. The Cityscapes also offers a wide range of design choices, including the following:
- Polygonal annotations
- Dense semantic segmentation
- Instance segmentation for vehicle and pedestrians
Here 30 classes are distributed into the 8 super categories:
- Flat surfaces
With the availability of 5,000 perfectly annotated images and an extra batch of 20,000 coarse annotations, you can apply the dataset for academic and non-academic purposes, as long as you seek non-commercial usage. The Cityscapes especially supports the use of the data for teaching, research, scientific publications, and personal experimentations.
The dataset is best for assessing the vision algorithms on a pixel, instance-level, panoptic-level, as well as for supporting research that is invested in large volumes of data. The annotations cover data recorded on different months to cover all seasons, daytime and nighttime, different variations of scene layout, and backgrounds. The Cityscapes dataset also provides GPS coordinates, outside temperature, ego-motion data, and right stereo views.
Annotation Type: instance segmentation
Created By: Cityscapes
Publish Date: February 20, 2016
Dataset Size: 5,000 images