CaiRoad:

The direct download link for the data can be found here https://www.cs.cornell.edu/projects/satellite-change-events/static/data/CaiRoad.zip.

Once unzipped the directory has the following structure.

CaiRoad/
    events/
        2955_3125_00001/
             00_img.png
             00_mask.png
             01_img.png
             ....
        2955_3125_00003/
        ...
    fulldata/
        2955_3125/
             bbox31250_29550_2015_07.jpg
             bbox31250_29550_2015_08.jpg
              ...
        2955_3130/
        ...
    road_labels.csv
    testset.csv
    trainset.csv

We describe each of these directories and files here:

fulldata: contains a set of directories for different locations. 
    Each directory name describes the bottom left corner of the bounding box region in coordinates. 
    If the directory name is  2955_3125. The bottom left corner has the coordinate (29.55 N 31.25 E) i.e. a division by 100. Each directory contains images from different months. 
    The last two integers delimited by an underscore decribe the year and month of the image. For example, *_2015_07.jpg was taken in July of 2015.
    The images itself is a 1024x1024 RGB representation of the signals from satellite capturing 0.05 degrees of latitute and longitude.

events: contains a set of directories for different events discovered and presented by in dataset.
    Each directory name is a unique identifier for an event.
    The event directory contains n images (**_img.png) and n-1 masks (**_mask.png). 
    The img.png files contain the RGB images from change events.
    i_mask.png is a binary mask showing change between i_img.png and i+1_img.png.

road_labels.csv: is a csv file containing ground truth labels for all the change events.
    Each row has 2 items, the first is the unique indentifier for events. 
    second is a 0, 1 value. If the value is 1, humans annotated is as a road construction event.

testset.csv, trainset.csv: are 50-50 test-train split for change events.
    the structure is similar to road_labels.csv
-----------------------------------------------------------------------------------------

CalFire:

The direct download link for the data can be found here https://www.cs.cornell.edu/projects/satellite-change-events/static/data/CalFire.zip.

The structure is very similar to CaiRoad.

Once unzipped the directory has the following structure.

CalFire/
    events/
        0089d3c4-c554-46e1-8eb9-af95e4dfae0c_00006/
             00_img.png
             00_mask.png
             01_img.png
             ....
        0089d3c4-c554-46e1-8eb9-af95e4dfae0c_00009/
        ...
    fulldata/
        0089d3c4-c554-46e1-8eb9-af95e4dfae0c/
             bbox_2019_04.jpg
             bbox_2019_05.jpg
              ...
        00a13d84-415e-4106-aa31-2e664ed9eabf/
        ...
    fire_labels.csv
    testset.csv
    trainset.csv

We describe each of these directories and files here:

fulldata: contains a set of directories for different locations. 
    Each directory name is the unique forest fire location (see forest unique identifier in 'all_forest_fires.csv'). 
    Some of these fires are small in area and temporal extent and are thus ignored.
    More information about obtaining images is provided with the code.

events: contains a set of directories for different events discovered and presented by in dataset.
    Each directory name is a unique identifier for an event.
    The event directory contains n images (**_img.png) and n-1 masks (**_mask.png). 
    The img.png files contain the RGB images from change events.
    i_mask.png is a binary mask showing change between i_img.png and i+1_img.png.

fire_labels.csv: is a csv file containing ground truth labels for all the change events.
    Each row has 2 items, the first is the unique indentifier for events. 
    second is a 0, 1 value. If the value is 1, it is annotated is as a forest fire event.

testset.csv, trainset.csv: are 50-50 test-train split for change events.
    the structure is similar to road_labels.csv