cub_get_database.m

function imdb = cub_get_database(cubDir, useCropped)

if (nargin < 2)
    useCropped = false;
end
% Automatically change directories
if useCropped
    imdb.imageDir = fullfile(cubDir, 'images_cropped') ;
else
    imdb.imageDir = fullfile(cubDir, 'images');
end

imdb.maskDir = fullfile(cubDir, 'masks'); % doesn't exist
imdb.sets = {'train', 'val', 'test'};

% Class names
[~, classNames] = textread(fullfile(cubDir, 'classes.txt'), '%d %s');
imdb.classes.name = horzcat(classNames(:));

% Image names
[~, imageNames] = textread(fullfile(cubDir, 'images.txt'), '%d %s');
imdb.images.name = imageNames;
imdb.images.id = (1:numel(imdb.images.name));

% Class labels
[~, classLabel] = textread(fullfile(cubDir, 'image_class_labels.txt'), '%d %d');
imdb.images.label = reshape(classLabel, 1, numel(classLabel));

% Bounding boxes
[~,x, y, w, h] = textread(fullfile(cubDir, 'bounding_boxes.txt'), '%d %f %f %f %f');
imdb.images.bounds = round([x y x+w-1 y+h-1]');

% Image sets
[~, imageSet] = textread(fullfile(cubDir, 'train_test_split.txt'), '%d %d');
imdb.images.set = zeros(1,length(imdb.images.id));
imdb.images.set(imageSet == 1) = 1;
imdb.images.set(imageSet == 0) = 3;

% Write out the segments
imdb.segments = imdb.images;
imdb.segments.imageId = imdb.images.id;
imdb.segments.mask = strrep(imdb.images.name, 'image', 'mask');

% make this compatible with the OS imdb
imdb.meta.classes = imdb.classes.name ;
imdb.meta.inUse = true(1,numel(imdb.meta.classes)) ;
imdb.segments.difficult = false(1, numel(imdb.segments.id)) ;