IDEA: Image database for earthquake damage annotationZenodo

The data article presents the “Image Database for Earthquake damage Annotation (IDEA)”, an extended dataset of annotated real structural damage consisting of more than 5400 images, collected during post-event and ordinary field inspections. The dataset aims to fill the lack of annotated data necessa...

Full description

Saved in:
Bibliographic Details
Main Authors: Ilaria Senaldi, Chiara Casarotti, Martina Mandirola, Alessio Cantoni
Format: Article
Language:English
Published: Elsevier 2025-08-01
Series:Data in Brief
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2352340925004603
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The data article presents the “Image Database for Earthquake damage Annotation (IDEA)”, an extended dataset of annotated real structural damage consisting of more than 5400 images, collected during post-event and ordinary field inspections. The dataset aims to fill the lack of annotated data necessary for the development of deep learning methodologies with structural damage detection and/or classification purposes. The dataset contains images annotated by structural engineers, covering different structural typologies, construction materials and damage typologies. The dataset is based on a comprehensive ontology defined by the authors, based on commonly agreed structural damage categories, which includes several types of structural and non-structural damage. Such onthology, can be used either to expand the presented dataset or to produce new ones, in order to increase the availability of data annotated according to a common standard, from the structural engineering point of view. Furthermore, the IDEA dataset is valuable as benchmark for enhancing the performance of damage classification/detection algorithms, encompassing some of the limits of currently available datasets, which cover only a few structural typologies or damage classes, or consist of classified rather than annotated images, or originate from limited laboratory experiments rather than post-event reconnaissance.
ISSN:2352-3409