This archive contains the data set used in the paper

Gencer Sumbul, R. Gokberk Cinbis, Selim Aksoy, "Fine-Grained Object Recognition
and Zero-Shot Learning in Remote Sensing Imagery," IEEE Transactions on
Geoscience and Remote Sensing, vol. 56, no. 2, pp. 770-779, February 2018.
DOI: 10.1109/TGRS.2017.2754648

If you use this data set in your research, we appreciate if you cite the above
paper.

1. OPENING THE DATA FILE (data.mat):

In order to load the dataset in MATLAB, you can use

load data;

For Python, you can use the following code snippet

import scipy.io as sio
data = sio.loadmat('data.mat')

2. CONTENT OF THE DATA SET:

Loading in MATLAB results in two cell arrays: data and classes. The first one,
data, is a 48063 x 2 cell array in which each row corresponds to an instance.
The first entry (column) for each instance is an image patch of size 25 x 25 x 3
pixels. The second entry is the class id of that instance. The second cell
array, classes, has size 40 x 3. Each row corresponds to a class. The first
entry for each class is a string that represents the common name of that class.
The second entry is a string that represents the scientific name of the class.
The third entry is the corresponding class embedding vector. The row indices
correspond to the class ids in the first cell array, data.

For Python, the resulting variable, data, is a dictionary. data['data'] is
a 48063 x 2 numpy array in which each row corresponds to an instance.
The first entry for each instance is an image patch of size 25 x 25 x 3
pixels as a numpy array. The second entry is the class id of that instance.
data['classes'] is a 40 x 3 numpy array. Each row corresponds to a class.
The first entry for each class is a numpy array with shape (1,) that represents
the common name of that class. The second entry is also a numpy array with
shape (1,) that represents the scientific name of the class. The third entry
is the corresponding class embedding vector that is a numpy array with shape
(1, 1150). The row indices + 1 correspond to the class ids.
