kevincluo/structure_wildfire_damage_classification

Dataset Card for Structures Damaged by Wildfire

Homepage: Image Dataset of Structures Damaged by Wildfire in California 2020-2022

Dataset Summary

The dataset contains over 18,000 images of homes damaged by wildfire between 2020 and 2022 in California, USA, captured by the California Department of Forestry and Fire Protection (Cal Fire) during the damage assessment process. The dataset spans across more than 18 wildfire events, including the 2020 August Complex Fire, the first recorded "gigafire" event in California where the area burned exceeded 1 million acres. Each image, corresponding to a built structure, is classified by government damage assessors into 6 different categories: Inaccessible (image taken but no assessment made), No Damage, Affected (1-9%), Minor (10-25%), Major (26-50%), and Destroyed (>50%). While over 57,000 structures were evaluated during the damage assessment process, only about 18,000 contains images; additional data about the structures, such as the street address or structure materials, for both those with and without corresponding images can be accessed in the "Additional Attribute Data" file.

The 18 wildfire events captured in the dataset are:

• [AUG] August Complex (2020)
• [BEA] Bear Fire (2020)
• [BEU] BEU Lightning Complex Fire (2020)
• [CAL] Caldor Fire (2021)
• [CAS] Castle Fire (2020)
• [CRE] Creek Fire (2020)
• [DIN] DINS Statewide (Collection of Smaller Fires, 2021)
• [DIX[ Dixie Fire (2021)
• [FAI] Fairview Fire (2022)
• [FOR] Fork Fire (2022)
• [GLA] Glass Fire (2020)
• [MIL] Mill Mountain Fire (2022)
• [MON] Monument Fire (2021)
• [MOS] Mosquito Fire (2022)
• [POST] Post Fire (2020)
• [SCU] SCU Complex Fire (2020)
• [VAL] Valley Fire (2020)
• [ZOG] Zogg Fire (2020)

The author retrieved the data, originally published as GIS features layers, from from the publicly accessible CAL FIRE Hub, then subsequently processed it into image and tabular formats. The author collaborated with Cal Fire in working with the data, and has received explicit permission for republication.

Data Fields

The data instances have the following fields:

• image: A PIL.Image.Image object containing the image. Note that when accessing the image column: dataset[0]["image"] the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the "image" column, i.e. dataset[0]["image"] should always be preferred over dataset["image"][0].
• labels: an int classification label.

Class Label Mappings:

{
  "affected": 0,
  "destroyed": 1,
  "inaccessible": 2,
  "major": 3,
  "minor": 4,
  "no_damage": 5,
}

Data Splits

	train
# of examples	18,714