About TwineTwine NetworkContact
CityScapes Dataset
Cityscapes is a large-scale urban street-scene dataset with stereo video and high-quality pixel-level annotations, built for benchmarking semantic segmentation, instance segmentation, and panoptic scene understanding for autonomous driving and smart-city computer vision.
Files
25000
Size
51.92GB
Format
jpeg
Duration
Country
Worldwide
Participants
Languages
Updated
December 11, 2025
Description
The Cityscapes Dataset provides a diverse collection of stereo video sequences captured across 50 cities, representing real world urban variability in architecture, traffic density, weather, and lighting. It includes 5,000 finely annotated frames with pixel-accurate labels plus an additional 20,000 weakly annotated frames, enabling both high-fidelity evaluation and large-scale learning workflows.
Cityscapes is purpose-built for semantic urban scene understanding, supporting core vision tasks such as pixel-level semantic labeling, instance-level segmentation, and panoptic segmentation, key building blocks for autonomous driving perception stacks, robot navigation, and road-scene analytics. With its mix of fine and weak supervision, the dataset is also ideal for modern AI research themes like semi-supervised learning, weak supervision, self-training, and generalization testing of deep neural networks on complex street environments.
Dataset Technical Specification
Number of files:
25000
Total dataset size:
51.92GB
Duration:
Format:
jpeg
Sample rate:
Resolution:
Dataset Demographics
📍 Country:
Worldwide
🧍 Gender:
📅 Age:
👥 Number of participants:
🛡️ Consent & Compliance
This Cityscapes Dataset is made freely available to academic and non-academic entities for non-commercial purposes such as academic research, teaching, scientific publications, or personal experimentation.
