The purpose of this task is to identify images useful for humanitarian response. We have listed below several humanitarian categories to choose from. We will use these human-tagged images to train machine learning classifiers to automatically classify thousands of images posted on Twitter during disasters to help humanitarian organizations. Specifically, the task is, given an image, categorize it into one of the following categories. In addition, for some images, humanitarian organizations are also interested to understand the severity of the damages caused by the disaster events. In those cases, we need to categorize one of the damage severity labels.
The purpose of this task is to understand the type of information shared in an image, which was collected from Twitter during a disaster event.
Affected, injured, or dead people: if the image shows injured, dead, or affected people such as people in shelter facilities, sitting or lying outside, etc.
Infrastructure or utility damage: if the image shows any built structure affected or damaged by the disaster. This includes damaged houses, roads, buildings; flooded houses, streets, highways; blocked roads, bridges, pathways; collapsed bridges, power lines, communication poles, etc.
Natural landscape damage: if the image shows damage to the natural landscape such as flooded agricultural landscape, burning crop fields, fires in forests, uprooted trees, etc.
Vehicle damage: if the image shows any type of damaged vehicles such as cars, trucks, buses, motorcycles, boats, ships, trams, trains, helicopters, airplanes, etc.
Rescue and response efforts: if the image shows any type of rescue, volunteering, or response effort such as people being transported to safe places, people being evacuated from the hazardous area, people receiving medical aid or food, donation of money, blood, or services, etc.
Other relevant information: if the image does not belong to any of the above categories, but it still contains important information useful for humanitarian aid, then select this category. For instance, injured animals, maps and infographics related to the disaster, emergency or helpline numbers.
Not humanitarian: if the image is not relevant or useful for humanitarian aid and response such as non-disaster scenes, cartoons, advertisement banners, celebrities, etc.
Don't know or can't judge: if you don’t know or can't judge, for example, due to low image quality or any other issue.
If you labeled an image as one of these three damage-related categories, (i) “Infrastructure or utility damage”, (ii) “Natural landscape damage”, or (iii) “Vehicle damage”, then we further ask you to specify the severity of damage shown in the image. The damage severity levels are described as follows:
Severe: Substantial destruction of an infrastructure belongs to the severe damage category. For example, a non-livable or non-usable building, a non-crossable bridge, or a non-drivable road, destroyed, burned crops, forests are all examples of severely damaged infrastructures.
Mild: Partially destroyed buildings, bridges, houses, roads belong to mild damage category.
Little-to-none: Images that show damage-free infrastructure (except for wear and tear due to age or disrepair) belong to the little-or-no-damage category.
Don't know or can't judge: If you cannot determine the level of damage.