BowFire is a dataset of still images for evaluating fire detection. It contains 226 full‐scene images, of which 119 show fire and 107 are non‐fire. The fire images capture a variety of emergency scenarios (buildings, industrial sites, car accidents, riots), while the non‐fire set includes scenes with fire‐like color regions (e.g. sunsets, red/yellow objects) so as to challenge false‐positives.
Citation: Chino, D. Y., Avalhais, L. P., Rodrigues, J. F., & Traina, A. J. (2015, August). Bowfire: detection of fire in still images by integrating pixel color and texture analysis. In 2015 28th SIBGRAPI conference on graphics, patterns and images (pp. 95-102). IEEE. See original paper.
Arboretum is a C++ open-source software library developed by the Databases and Images Group at ICMC-USP that implements Metric Access Methods (MAMs) and provides an extensible platform for building content-based retrieval systems. It is organized in three layers: a User layer (where one defines object types and the distance/dissimilarity functions), a Structure layer (implementing metric indexing structures such as Slim-Tree and others), and a Storage layer (which handles paging and the storage of data/indexes on disk or in memory). Arboretum supports many distance functions (e.g. from the Minkowski family) and metric structures, allowing combinations (e.g. Slim-Tree with Euclidean, or with City-Block etc.), so that one can tailor both feature extraction and similarity measure to the domain.