pytorch implementation of network-in-network model on cifar10. All settings are base on the network-in-network model in Caffe model zoo.

$ git clone https://github.com/jiecaoyu/pytorch-nin-cifar10.git
$ cd pytorch-nin-cifar10
$ mkdir data

Then download the data from this link and uncompress it into the ./data/ directory. Now you can train the model by running

$ python original.py

By tweaking hyper-parameters, the model can reach the accuracy of 89.64%, which is better than other available Torch/PyTorch implementations.

The data used to train this model comes from http://www.cs.toronto.edu/~kriz/cifar.html Please follow the license there if used.