March 29, 2018

Announcing the DeepGlobe Satellite Challenge for CVPR 2018

By: Ilke Demir, Manohar Paluri, Ramesh Raskar

We are excited to announce the DeepGlobe Satellite Challenge, for the DeepGlobe workshop happening at CVPR 2018.

The workshop, which is a collaborative effort from Facebook, DigitalGlobe, Purdue, and MIT, aims to bring together a diverse set of researchers to advance the state-of-the-art in satellite image analysis by providing datasets and a competition platform to host and evaluate computer vision and machine learning solutions. The challenge is structured around three different satellite image understanding tasks, road extraction, building detection, and land cover classification.

What is DeepGlobe?

We observe that satellite imagery is a powerful source of information as it contains more structured and uniform data, compared to traditional images. Although the computer vision community has been accomplishing hard tasks on everyday image datasets using deep learning, satellite images are only recently gaining attention for maps and population analysis.

To direct more attention to satellite image approaches, we are announcing the DeepGlobe Satellite Image Understanding Challenge. We expect that the datasets created and released for this competition may serve as reference benchmarks for future research in satellite image analysis. And, because the challenge tasks will involve “in the wild” forms of classic computer vision problems, the datasets have the potential to become valuable testbeds for the design of robust vision algorithms, beyond the area of remote sensing.

The challenge

Challengers will be provided with high-resolution satellite image datasets (courtesy of DigitalGlobe) and the corresponding training data. We expect them to learn the expected urban elements for each category: road extraction, building detection and land cover classification. Live scores, submission and evaluation of the results, and the datasets will be maintained in the workshop website. Challengers will also be required to submit a short paper (up to 4 pages) detailing their methodology, which can be extended as a full paper for further publication. The challenge is underway now, results and the short papers must be submitted by May 1st.

For more information visit the challenge website: