MXNET is a great deep learning framework, and currently has the best support for distribued trainning. However, its documentation lacks maintaince and I get confused in many API details. Though MXNET community has provided a bunch of examples, I do not find useful examples for cifar 100. Since MXNET is using its own data format rec
, I have to convert data to rec
before training. data.mxnet.io/data only provides rec files of cifar 10. It takes some time for me to figure out how to generate my own cifar 100 rec file.
First, download python version of cifar 100
Then, dump it into images, thanks to apache/mxnet#235
import os
import sys
import pickle
import csv