Building Missing Maps with Machine Learning
41 days left
We are in a period of increasing humanitarian crises, both in scale and number. Natural disasters continue to increase in frequency and impact, while long-term and reignited conflicts affect people in many parts of the world. Often, accurate maps either do not exist or are outdated by disaster or conflict.
Humanity & Inclusion is an aid organization working in some 60 countries, alongside people with disabilities and vulnerable populations. Our emergency sector responds quickly and effectively to natural and civil disasters.
Many parts of the world have not been mapped; especially the most marginalized parts, that is, those most vulnerable to natural hazards. Obtaining maps of these potential crisis areas greatly improves the response of emergency preparedness actors.
During a disaster it is extremely useful to be able to map the impassable sections of road for example, as well as the most damaged residential areas, the most vulnerable schools and public buildings, population movements, etc. The objective is to adapt as quickly as possible the intervention procedures to the evolution of the context generated by the crisis.
In the first days following the occurrence of a disaster, it is essential to have as fine a mapping as possible of communication networks, housing areas and infrastructures, areas dedicated to agriculture, etc.
Today, when new maps are needed they are drawn by hand, often by volunteers who participate in so called Mapathons. They draw roads and buildings on satellite images, and contribute to Open StreetMap.
For instance, Humanity & Inclusion has been involved in organising numerous Mapathons to draw new maps for our clearance teams in Laos.
In this challenge we want to explore how Machine Learning can help pave the way for automated analysis of satellite imagery to generate relevant and real-time maps.
Satellite imagery is readily available to humanitarian organisations, but translating images into maps is an intensive effort. Today maps are produced by specialized organisations or in volunteer events such as mapathons, where imagery is annotated with roads, buildings, farms, rivers etc.
Images are increasingly available from a variety of sources, including nano-satellites, drones and conventional high altitude satellites. The data is available: the task is to produce intervention-specific maps with the relevant features, in a short timeframe and from disparate data sources.
In this challenge you will be provided with a dataset of individual tiles of satellite imagery as RGB images, and their corresponding annotations of where an image is there a building. The goal is to train a model which given a new tile can annotate all buildings.
Also, in context of this challenge, to make the barrier to entry much lower, we tried to remove all the domain specific jargon of Remote Sensing and Satellite Imagery Analysis, and are presenting this as a problem of Object Detection and Object Segmentation in Images.
The idea being, once we collectively demonstrate that an approach works really well on RGB images with just 3 channels of information, we can then work on extending it to multi-channel information from rich satellite imagery.
You can download the datasets in the Datasets Section. You are provided with :
train.tar.gz: This is the Training Set of 280741 tiles (as 300x300 pixel RGB images) of satellite imagery, along with their corresponding annotations in MS-COCO format
val.tar.gz: This is the suggested Validation Set of 60317 tiles (as 300x300 pixel RGB images) of satellite imagery, along with their corresponding annotations in MS-COCO format
test_images.tar.gz: This is the Test Set for Round-1, where you are provided with 60697 files (as 300x300 pixel RGB images) and your are required to submit annotations for all these files.
For more details about the dataset, and submission procedures etc, please refer to the following notebooks :
- Dataset Utils
- Random Submission
- Train Mask-RCNN
For for a known ground truth mask , you propose a mask , then we first compute (Intersection Over Union) :
measures the overall overlap between the true region and the proposed region. Then we consider it a True detection, when there is atleast half an overlap, or when
Then we can define the following parameters :
The final scoring parameters and are computed by averaging over all the precision and recall values for all known annotations in the ground truth.
We will stop accepting submissions for Round 1 on June 1, 2018. The top 30 participants of Round 1 will be invited to complete in Round 2.
For instructions on submitting solutions for Round-2, please refer to the mapping-challenge-starter-kit.
Note: We will be adding more content to the starter kit to help you get started in the challenge. So please do keep a close eye on the starter-kit for updates. In the meantime, you can have a look at the examples of cocoapi on how to easily parse and explore the datasets.
The top scoring 30 participants in Round 1 will be invited to participate in round 2. Participants will be required to submit their code and models which will be internally tested by UNOSAT on a dataset (in a similar format as the currently released data) of an undisclosed location.
Detailed instructions around how to package your code and models for submissions will be announced over the next few months.
Round 1 : 28.03.2018 - 01.06.2018
Round 2 : 02.06.2018 - 15.07.2018
Announcement of Overall Results : 20.07.2018
Tech4Dev 2018 International Conference
The UNESCO Chair in Technologies for Development hosted by the Cooperation & Development Center (CODEV) at the École Polytechnique Fédérale de Lausanne (EPFL) in Switzerland is pleased to announce the 5th International Conference on Technologies for Development (Tech4Dev 2018) taking place from 27-29 June 2018 at the SwissTech Convention Center in Lausanne, Switzerland.
The Tech4Dev Conference is the biennial flagship event of the UNESCO Chair in Technologies for Development hosted by CODEV at EPFL. The Conference focuses on the potential of technology solutions to advance inclusive social and economic development in the Global South.Tech4Dev 2018 puts the challenges and the potential of the Global South in the center of discussions reflected by its title Voices of the Global South. Indeed, even though the Sustainable Development Goals apply globally, the needs in terms of access to health and financial services, functioning education systems and sustainable urban and natural ecosystems are undeniably more significant in the Global South. In this perspective, it is crucial to listen to and support stakeholders from the regions who are facing these challenges.
Winners of Round-1 of the Mapping Challenge will be invited, with travel and accommodation paid, to demonstrate and present their solution for the challenge within the Global Humanitarian Lab exhibition space.
The 2018 International Tech4Dev Conference is from 27-29th June, 2018.
The following rules have to be observed by all participants:
- Participants are allowed at most 5 submissions per day.
- Participants are welcome to form teams. Teams should submit their predictions under a single account. The submitted paper will mention all members.
- Participants have to release their solution under an Open Source License of their choice to be eligible for prizes. We encourage all participants to open-source their code!
- Participants are not allowed to use any other datasets other than the ones released in context of this challenge. The use of pre-trained models is nevertheless permitted.
- While submissions by Admins and Organizers can serve as baselines, they won’t be considered in the final leaderboard.
- In case of conflicts, the decision of the Organizers will be final and binding.
- Organizers reserve the right to make changes to the rules and timeline.
- Violation of the rules or other unfair activity may result in disqualification.
- Round 1 : Invitation to present the winning solution at the Global Humanitarian Lab’s exhibit space at the 2018 Tech4Dev Event, with travel and accommodation covered.
- Round 2 : Invitation to the 3rd Applied Machine Learning Days at EPFL in Switzerland in January 2019, with travel and accommodation covered.
- Round 1 + Round 2 : A printed award of appreciation from the Global Humanitarian Lab, Handicap International and CartONG.
A starter kit has been prepared which explains all the nuts and bolts required to get started in the challenge. It can be accessed at : https://github.com/crowdAI/mapping-challenge-starter-kit
- Gitter Channel : crowdAI/crowdai-mapping-challenge
- Technical issues : https://github.com/crowdAI/mapping-challenge-starter-kit/issues
- Discussion Forum : https://www.crowdai.org/challenges/mapping-challenge/topics
We strongly encourage you to use the public channels mentioned above for communications between the participants and the organizers. In extreme cases, if there are any queries or comments that you would like to make using a private communication channel, then you can send us an email at :