PEER Hub ImageNet (PHI) Challenge Pre-Announcement

April 12, 2018

2018 Challenge image

The Pacific Earthquake Engineering Research Center (PEER) is organizing the first image-based structural damage identification competition, namely PEER Hub ImageNet (PHI) Challenge to be announced in the mid-summer of 2018.

In this challenge, two sets of images will be provided to the contestants, one for training and the other for testing. The first set consists of about 20,000 labeled images for different categories, examples of which are structural component type, damage level, and damage type. Each competing team is expected to use/develop algorithms to train their recognition models based on these well-labeled images. The second set consists of 5,000 unlabeled images to be labeled by the teams using their trained algorithms. Labels predicted for the test set will be compared against reference labels and teams with the highest accuracy will be declared the winners of the challenge. Reference labels will be provided by a team of structural experts determined by the competition organization committee. Scoring and other rules will be provided during the formal challenge announcement.

This effort is part of PEER’s strategic plan of equipping the earthquake engineering community with tools of the current Digital Revolution Era of Machine Learning, Deep Learning, Artificial Intelligence, and High-Performance Computing. The objective of this challenge is to fully engage the earthquake engineering and other extreme events community at all stages, including preparation of the datasets, execution of the computations, and processing and interpretation of the results. In the datasets preparation stage, members of the community are expected to contribute by uploading images that can be used in the challenge and by labeling these images. The staff of the PEER Center has developed apps for uploading and labeling images, which can be found in this link.

Please, visit the website to contribute to the challenge preparation and development of a large structural image dataset.

The announcement of the challenge, including complete list of its rules, will be released around mid-June 2018. The competition website is under preparation and will be accessible soon.