HRSCD - High Resolution Semantic Change Detection Dataset

Citation Author(s):
Caye Daudt
ETH Zürich
Le Saux
Télécom Paris
Submitted by:
Rodrigo Daudt
Last updated:
Thu, 12/15/2022 - 11:33
9 ratings - Please login to submit your rating.




Dataset described in: 

Daudt, R.C., Le Saux, B., Boulch, A. and Gousseau, Y., 2019. Multitask learning for large-scale semantic change detection. Computer Vision and Image Understanding, 187, p.102783.


This dataset contains 291 coregistered image pairs of RGB aerial images from IGS's BD ORTHO database. Pixel-level change and land cover annotations are provided, generated by rasterizing Urban Atlas 2006, Urban Atlas 2012, and Urban Atlas Change 2006-2012 maps. 


The dataset is split into five parts:

    - 2006 images 

    - 2012 images

    - Change labels

    - 2006 land cover maps

    - 2012 land cover maps




Change labels are available for all image pairs in the dataset. Available classes:

    - 0: No change

    - 1: Change


Land cover maps are available for all images in the dataset. Urban Atlas classes have been grouped at first hierarchical level, as described in the paper cited above. Available classes:

    - 0: No information

    - 1: Artificial surfaces

    - 2: Agricultural areas

    - 3: Forests

    - 4: Wetlands

    - 5: Water




If you use this dataset for your work, please use the following citation:



title = "Multitask Learning for Large-scale Semantic Change Detection",

journal = "Computer Vision and Image Understanding",

volume = "187",

pages = "102783",

year = "2019",

issn = "1077-3142",

doi = "",

url = "",

author = "Daudt, {Rodrigo Caye} and {Le Saux}, Bertrand and Boulch, Alexandre and Gousseau, Yann",

keywords = "Semantic change detection, High resolution Earth observation, Fully convolutional networks, Remote sensing, Multitask learning"




Rodrigo Caye Daudt,

Bertrand Le Saux,

Alexandre Boulch,

Yann Gousseau,





The images in this dataset are released under IGN's "licence ouverte". More information can be found at


The maps used to generate the labels in this dataset come from the Copernicus program, and as such are subject to the terms described here:


This dataset is released under Creative-Commons BY-NC-SA licence. For commercial purposes, please contact the authors.



Please contact us if you have any questions.


The - BD ORTHO® 50 cm, D 14 - CALVADOS, 2005 and  - BD ORTHO® 50 cm, D 35 - ILLE-ET-VILLAINE, 2006 datasets do not appear on the linked IGN webpage provided.  Additionally the Geoportail provided is difficult to navigate for those of us who don't speak French.  Is there an updated webpage I can access these datasets at?  Thank you in advance.

Submitted by Austin Willoughby on Thu, 06/11/2020 - 12:01

As stated in the dataset description, we do not have the redistribution rights for the 2006 images. Please contact us by email for further help in downloading these images.

Submitted by Rodrigo Daudt on Fri, 07/03/2020 - 03:53

Have you got the image data for 2006?

Submitted by qiu ru ren on Mon, 07/06/2020 - 06:16

The 2006 images appear to be missing

Submitted by Shairoz Sohail on Thu, 07/02/2020 - 19:30

As stated in the dataset description, we do not have the redistribution rights for the 2006 images. Please contact us by email for further help in downloading these images.

Submitted by Rodrigo Daudt on Fri, 07/03/2020 - 03:53


Submitted by Rodrigo Daudt on Fri, 07/03/2020 - 03:54

All change labels files appear to be blank or else are only comprised of 0's (i.e. "no change").  Is there a way to fix this?

Submitted by Austin Willoughby on Tue, 07/21/2020 - 10:41

Some people have contacted me with this problem, and it seems that OpenCV sometimes fails to load the labels correctly, I don't know why. Below is a python program you can run to test if the images are corrupted. The output should contain mostly ones and a few zeroes.



import as io

from glob import glob

files = glob('change/*/*.tif')

for f in files:

    I = io.imread(f)


Submitted by Rodrigo Daudt on Tue, 07/21/2020 - 12:56

Using This dataset, is their a sample project available for semantic change detection. Kindly share any github page, that would be helpful for my research work. 

Submitted by Visalakshi kana... on Thu, 07/23/2020 - 15:34

Hi Rodrigo,

The datset looks not correct. The 2006 & 2012 Lancover maps are missing. Also Tiff images present in 2006/2012/change folders all looks simillar. Can you provide the correct dataset and GIThub page Simillar to OSCD dataset. 



Submitted by Visalakshi kana... on Tue, 07/28/2020 - 13:45

Totally agree!!

Submitted by Wentao Luo on Sun, 10/11/2020 - 07:53

Thank you for upload. However the "images 2012" zip file consists of 291 single images not pair. Is there anything wrong?

Submitted by sami temeltas on Fri, 10/16/2020 - 15:39

Unique labels for all the land cover maps show 0 instead of 0,1,2,3,4,5. Can you please guide me on this

Submitted by Anisha Chakravorty on Fri, 12/16/2022 - 06:09

The 2006 images appear to be missing, and I've contacted you by e-mail, please check.

Submitted by Jiahao Wang on Fri, 12/22/2023 - 19:55

Dataset Files

Open Access dataset files are accessible to all logged in  users. Don't have a login?  Create a free IEEE account.  IEEE Membership is not required.


File README.txt3.26 KB