Data Description
This is the CIFAR-10 Dataset consists of a collection of 60,000 images of 32x32 pixels with three 8-bit RGB channels.   Note that in some machine learning frameworks this data set can be downloaded directly from into data structures useful for processing in the CNN.  You are welcome to use those APIs to access the data set as an alternative to manually downloading the data posted here.  Be sure to keep the last 10,000 images as your test dataset for submitting predictions.
To read in Python a batch of images from the data files posted here, please use the following approach.  For python 2, you can use the cPickle module to the load the batch of data as
import cPickle
def loadImageBatch(filename):
fid = open(filename, 'rb');
data_batch = cPickle.load(fid);
return data_batch;
For python 3, you can use pickle module to load the batch of data as
import pickle
def loadImageBatch(filename):
fid = open(filename, 'rb');
data_batch = pickle.load(fid,encoding='bytes');
return data_batch;
The data batches are numpy arrays of dimension 10000x3072 of data type uint8.   Each image can be obtained by reshaping the 3072 vector into a 32x32x3 array encoding the RGB image.  More information can also be found on the CIFAR-10 dataset website, http://www.cs.toronto.edu/~kriz/cifar.html.
File descriptions
cifar-10-python.tar.gz
- the data_batchX with X in the range 1-5 are the training data images and test_batch is the test data images on which to make predictions.