The Cityscapes Dataset

By Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Scharwächter, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, Bernt Schiele. 2015

399 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

Citations in the corpus

Semantic Understanding of Scenes Through the ADE20K Dataset (2016, methodology): “However, the current datasets have limited number of objects (e.g., COCO [14], Pascal [9]) and in many cases those objects are not the most common objects one encounters in the world (like frisbees or baseball bats), or the datasets only cover a limited set of scenes (e.g., Cityscapes [6]).”
Unsupervised Image-to-Image Translation Networks (2017, methodology): “we showed several example results achieved by applying the proposed framework to translate images between the synthetic images in the SYNTHIA dataset [23] and the real images in the Cityscape dataset [2]. For the real to synthetic translation, we found our method made the cityscape images cartoon li”
A Simple Framework for Open-Vocabulary Segmentation and Detection (2023, methodology): “During fine-tuning, we use 512 × 1024 for Cityscapes [6] and 640 × 640 for ADE20K [58] by default.”
Generalized Decoding for Pixel, Image, and Language (2022, background): “In Table 2, we evaluate our model in a zero-shot manner on seven commonly used segmentation datasets in 10 different settings from diverse domains, including common indoor (e.g., ADE20K [98] and Pascal [22]), outdoor (e.g., Cityscapes [14]) and self-driving scenarios (e.g., BDD [90]).”
Scene Parsing through ADE20K Dataset (2017, methodology): “However, the current datasets have limited number of objects (e.g., COCO [17], Pascal [10]) and in many cases those objects are not the most common objects one encounters in the world (like frisbees or baseball bats), or the datasets only cover a limited set of scenes (e.g., Cityscapes [7]).”
Higher Order Conditional Random Fields in Deep Neural Networks (2015, methodology): “of superpixel-based potentials. We trained on the provided training set and evaluated our method on the validation set. 6.4. Cityscapes Dataset We nally evaluated our method on the Cityscapes dataset [5]. It contains images of street scenes from different cities with 20 ;000 coarse and 2;975 ne anno”
Higher Order Potentials in End-to-End Trainable Conditional Random Fields (2015, methodology): “We finally evaluated our method on the Cityscapes dataset [5].”
Adaptive VIO: Deep Visual-Inertial Odometry with Online Continual Learning (2024, neutral): “While these online learning methods have shown adaptability in autonomous driving scenarios such as Cityscape[lO] and KITTI [13], they often fail in complex environments and under complex motions.”
The SYNTHIA Dataset: A Large Collection of Synthetic Images for Semantic Segmentation of Urban Scenes (2016, background): “To date, the largest dataset for semantic segmentation is the CityScapes dataset [8], which consists of a collection of images acquired in 50 cities around Germany, Switzerland and France in different seasons, and having 5,000 images with fine annotations and 20,000 with coarse annotations over a…”
A survey on deep learning techniques for image and video semantic segmentation (2018, background): “• Cityscapes[36](9): is a large-scale database which focuses on semantic understanding of urban street scenes.”
Dynamic-Structured Semantic Propagation Network (2018, methodology): “Coco-Stuff [4], ADE20k [41], Cityscape [6] and Mapillary [27]) demonstrate the effectiveness of incorporating our DSSPN into the state-of-the-art basic segmentation networks.”
Semantic Instance Annotation of Street Scenes by 3D to 2D Label Transfer (2015, methodology): “s semantic information only for a few object categories1, CamVid [8] offers pixel-accurate labels, but without instances and for a very limited number of frames. Very recently, the Cityscapes dataset [11] has been proposed with 5k manually annotated individual 2D images of street scenes2. Our datase”
In Defense of Pre-Trained ImageNet Architectures for Real-Time Semantic Segmentation of Road-Driving Images (2019, methodology): “We conduct semantic segmentation experiments on two datasets: Cityscapes [5] and CamVid [1].”
Open Vocabulary Scene Parsing (2017, methodology): “These networks perform well on datasets like PASCAL VOC [8] with 20 object categories, Cityscapes [4] with 30 classes, and a recently released benchmark SceneParse150 [33] covering 150 most frequent daily objects.”
Cross-view Transformers for real-time Map-view Semantic Segmentation (2022, background): “Driven by ever larger 3D recognition datasets [3, 5, 10, 13, 38], a number of works have focused on perception in the map-view.”
Shunted Self-Attention via Multi-Scale Token Aggregation (2021, neutral): “…[6], object detec-* �0�R�G�H�O��6�L�]�H�� 0� �7 �R�S � � � �$ �F�F �X �U �D�F �\ � � � � �7�R�S��$�F�F�X�U�D�F�\��R�Q��,�P�D�J�H�1�H�W �5�H�V�1�H�W �'�H�L�7 �6�Z�L�Q �3�9�7 �)�R�F�D�O �2�X�U�V tion [8,13], semantic segmentation [4,36] and video action recognition [15, 22].”