Events Dataset for Image Sanitization.

Overview

Datasets introduced in "Semi-Supervised Feature Embedding for Data Sanitization in Real-World Events"

It includes the links for the images that consists the five datasets.

NotreDame: Remarkable partial destruction of the Parisian Cathedral by a fire in 2019.

Grenfell: 2017 tragic fire incident in the Grenfell Tower in London.

NationalMuseum: Total destruction of the National Museum in Brazil by flames in 2018.

BangladeshFire: Fast-moving fire in a district in Dakha, that took place in 2019.

BostonMarathon: 2013 terrorist attack on the traditional Bostonian event.

Observations:

- We did not publish the links for the positive images for Grenfell dataset due to copyrights reasons.

- BostonMarathon and Grenfell negative sets are pictures from Flickr100k dataset

- Since BostonMarathon positive samples contains frames from Youtube videos, we published the Youtube video URL and the number of the frame that was extracted.

- BostonMarathon positive sample contains augmentation, that are crops which size is half of the original image. In this sense, columns i and j indicates the upper-left pixel of the crop. For instance, if the image is a x b, the crop will be the square between the points: j x i; (j + a/2) x i; j x (i + b/2); and (j + a/2) x (i + b/2).

Media Content

Due to the terms of use from the social networks, we do not make publicly available the texts, images and videos that were collected (only the links). However, we can provide some extra piece of media content related to one (or more) events by contacting the authors.

Funding

DéjàVu thematic project, São Paulo Research Foundation (2017/12646-3, 2018/05668-3, and 2020/02241-9)