OSCD - Onera Satellite Change Detection

Citation Author(s):
Rodrigo
Caye Daudt
ONERA
Bertrand
Le Saux
ESA/ESRIN
Alexandre
Boulch
Valeo.ai
Yann
Gousseau
Telecom ParisTech
Submitted by:
Bertrand Le Saux
Last updated:
Sat, 06/06/2020 - 12:11
DOI:
10.21227/asqe-7s69
Data Format:
Links:
License:
4.5
4 ratings - Please login to submit your rating.

Abstract 

The Dataset

The Onera Satellite Change Detection dataset addresses the issue of detecting changes between satellite images from different dates.

It comprises 24 pairs of multispectral images taken from the Sentinel-2 satellites between 2015 and 2018. Locations are picked all over the world, in Brazil, USA, Europe, Middle-East and Asia. For each location, registered pairs of 13-band multispectral satellite images obtained by the Sentinel-2 satellites are provided. Images vary in spatial resolution between 10m, 20m and 60m.

Pixel-level change ground truth is provided for 14 of the image pairs. The annotated changes focus on urban changes, such as new buildings or new roads. These data can be used for training and setting parameters of change detection algorithms.

 

The Benchmark

The algorithms can be tested in a benchmark for change detection.

The ground truth for the 10 remaining images remain undisclosed. Change prediction maps can be uploaded for evaluation on the IEEE GRSS DASE website. Various metrics such as per-class accuracy and confusion matrices are automatically computed on the website, and are available for participants. Comparison to the best performing methods is provided in the leaderboard associated with this benchmark.

(Update June 2020) Alternatively, you can use the test labels which are now provided in a separate archive, and compute standard metrics. For this purpose, we provide you with a python notebook to train, apply and evaluate Fully-Convolutional Networks for change detection: https://github.com/rcdaudt/fully_convolutional_change_detection 

References

If you use this work for your projects, please take the time to cite our paper:

Urban Change Detection for Multispectral Earth Observation Using Convolutional Neural Networks, R. Caye Daudt, B. Le Saux, A. Boulch, Y. Gousseau. IEEE International Geoscience and Remote Sensing Symposium (IGARSS’2018). Valencia, Spain. July 2018.

[PDF] [BibTeX]

 

The Team

 

 

Instructions: 

Onera Satellite Change Detection dataset

 

##################################################

Authors: Rodrigo Caye Daudt, rodrigo.daudt@onera.fr

Bertrand Le Saux, bls@ieee.org

Alexandre Boulch, alexandre.boulch@valeo.ai

Yann Gousseau, yann.gousseau@telecom-paristech.fr

 

##################################################

About: This dataset contains registered pairs of 13-band multispectral satellite images obtained by the Sentinel-2 satellites of the Copernicus program. Pixel-level urban change groundtruth is provided. In case of discrepancies in image size, the older images with resolution of 10m per pixel is used. Images vary in spatial resolution between 10m, 20m and 60m. For more information, please refer to Sentinel-2 documentation.

 

For each location, folders imgs_1_rect and imgs_2_rect contain the same images as imgs_1 and imgs_2 resampled at 10m resolution and cropped accordingly for ease of use. The proposed split into train and test images is contained in the train.txt and test.txt files.

For downloading and cropping the images, the Medusa toolbox was used: https://github.com/aboulch/medusa_tb

For precise registration of the images, the GeFolki toolbox was used. https://github.com/aplyer/gefolki

 

##################################################

Labels: The train labels are available in two formats, a .png visualization image and a .tif label image. In the png image, 0 means no change and 255 means change. In the tif image, 0 means no change and 1 means change.

<ROOT_DIR>//cm/ contains: - cm.png - -cm.tif

Please note that prediction images should be formated as the -cm.tif rasters for upload and evaluation on DASE (http://dase.grss-ieee.org/).

(Update June 2020) Alternatively, you can use the test labels which are now provided in a separate archive, and compute standard metrics using the python notebook provided in this repo, along with a fulll script to train and classify fully-convolutional networks for change detection: https://github.com/rcdaudt/fully_convolutional_change_detection 

 

##################################################

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

@inproceedings{daudt-igarss18,

author = {{Caye Daudt}, R. and {Le Saux}, B. and Boulch, A. and Gousseau, Y.},

title = {Urban Change Detection for Multispectral Earth Observation Using Convolutional Neural Networks},

booktitle = {IEEE International Geoscience and Remote Sensing Symposium (IGARSS'2018)},

venue = {Valencia, Spain},

month = {July},

year = {2018},

}

 

##################################################

Copyright: Sentinel Images: This dataset contains modified Copernicus data from 2015-2018. Original Copernicus Sentinel Data available from the European Space Agency (https://sentinel.esa.int).

Change labels: Change maps are released under Creative-Commons BY-NC-SA. For commercial purposes, please contact the authors.

 

Comments

Hello, everyone. I uploaded my inference results on this page, but I can't see my results on the stat page.

Submitted by XIAOFENG JI on Tue, 05/26/2020 - 02:28

Dear XiaoFeng,

 

Thanks very much for you interest in our OSCD dataset.

We are sorry about the problems you encountered. We now provide an alternate solution:

Labels for test images have now been added to the dataset, and you can use scripts (look for the python notebook) in our github repository to compute evaluations:

https://github.com/rcdaudt/fully_convolutional_change_detection

 

Keep the good work going on,

Best regards,

Bertrand

Submitted by Bertrand Le Saux on Sat, 06/06/2020 - 12:17

Hi all,
Thank you so much to provide such a great labeled images! But could I ask a quick question? How did you guys label the images? through google earth? Do you guys have any other auxiliary sources for labelling?
Best,
Ming

Submitted by Yiming Zhang on Mon, 01/25/2021 - 09:58