Image database Print
Written by Damir Krstinic, Toni Jakovcevic   
Wednesday, 10 February 2010 13:01

Very important part in algorithm evaluations is the quality of testing images. Larger number of images containing different scenarios and surroundings gives the better quality of the evaluation process. Everybody is invited to download as well as contribute to this publicly available database.

Two image databases are formed. The first database contains a collection of non-segmented wildfire smoke images, photographed both from the ground and from the air. Sample images are shown below:

 


 

Registered users could upload their images too, so the idea is to create a comprehensive image base of wildfire smoke images available to all researchers.

The second database contains a selection of wildfire smoke images and the same images manually segmented by human reference observer in 3-classes defined as:


1.    smoke (white - #ffffff)
2.    maybe smoke (gray - #808080)
3.    no-smoke (black - #000000)


Image database is divided in 4 directories:


•    training (containing 49 original images with wildfire smoke)
•    gt_training (the same images as above manually segmented by human reference observer into three classes - smoke, maybe smoke and not smoke)
•    testing (containing another collection of 49 original images with wildfire smoke)
•    gt_testing (the same images as above manually segmented by human reference observer into three classes - smoke, maybe smoke and not smoke)

Sample images are shown below:

 


 

Users can download images from hereThese datasets are free to use, but if you intend to use them in scientific research, it is necessary to reference this webpage and the Center for Wildfire Research.

Last Updated on Monday, 22 January 2018 11:56