IEEE Transactions on Image Processing (TIP)

Night-time Scene Parsing with a Large Real Dataset

Xin Tan1,2     Ke Xu1,2     Ying Cao2     Yiheng Zhang3    
Lizhuang Ma1     Rynson W.H. Lau2    

1 Shanghai Jiao Tong University     2 City University of Hong Kong     3 Stanford University

Contact us:    tanxin2017 AT  
xtan AT  


Although huge progress has been made on scene analysis in recent years, most existing works assume the input images to be in day-time with good lighting conditions. In this work, we aim to address the night-time scene parsing (NTSP) problem, which has two main challenges: 1) labeled night-time data are scarce, and 2) over- and under-exposures may co-occur in the input night-time images and are not explicitly modeled in existing pipelines. To tackle the scarcity of night-time data, we collect a novel labeled dataset, named NightCity, of 4,297 real night-time images with ground truth pixel-level semantic annotations. To our knowledge, NightCity is the largest dataset for NTSP. In addition, we also propose an exposure-aware framework to address the NTSP problem through augmenting the segmentation process with explicitly learned exposure features. Extensive experiments show that training on NightCity can significantly improve NTSP performances and that our exposure-aware model outperforms the state-of-the-art methods, yielding top performances on our dataset as well as existing datasets. : [ NightCity.pdf ]


Images (1.8G) from both training set and testing set can be downloaded via [ Images-Google drive ] directly.
Labels (140MB) from both training set and testing set can be obtained via [ Labels-Google drive ]!
The reannotated val set can be downloaded at [ NightLab ]. (updated since 01, April, 2022)

Terms and Conditions
The dataset can be used freely if you agree with all the following terms.
- The dataset is used only for non-commercial purposes, such as teaching and research. You do not use the dataset or any of its modified versions for any purposes of commercial advantage or private financial gain.
- In case you use the dataset within your research papers, you refer to our publications on our website. If the dataset is used in media, a link to our website is included.
- We reserve all rights that are not explicitly granted to you. The dataset is provided as is, and you take full responsibility for any risk of using it. There may be inaccuracies although we tried, and will try our best to rectify any inaccuracy once found.


    author = {Tan, Xin and Xu, Ke and Cao, Ying and Zhang, Yiheng and Ma, Lizhuang and Lau, Rynson W. H.},
    title = {Night-time Scene Parsing with a Large Real Dataset},
    journal={IEEE Transactions on Image Processing},
    year = {2021},
    doi={10.1109/TIP.2021.3122004} }


Our method is trained on two 1080ti GPU cards with a small batch size. Welcome to see more comptetitive results with larger batch size.

Website visit statistics

This website is borrowed from Haiyang Mei.
Last updated in 1 April, 2022.