scarletclaw / resnet-tf

ResNet Implementation in TensorFlow

ResNet TensorFlow

This is a TensorFlow implementation of ResNet, a deep residual network developed by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.

Read the original paper: "Deep Residual Learning for Image Recognition" (http://arxiv.org/abs/1512.03385).

Disclaimer: I implemented this for only learning purposes. Check out the original repo for other unofficial implementations.

TODO:

• put CIFAR-10 data in a TensorFlow Dataset object

Getting Started

Cloning the repo

$ git clone http://github.com/xuyuwei/resnet-tf
$ cd resnet-tf

Setting up the virtualenv, installing TensorFlow (OS X)

$ virtualenv venv
$ source venv/bin/activate
(venv)$ pip install --upgrade https://storage.googleapis.com/tensorflow/mac/tensorflow-0.6.0-py2-none-any.whl

If you don't have virtualenv installed, run pip install virtualenv. Also, the cifar-10 data for python can be found at: https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz. Place the data in the main directory.

Start Training:

(venv)$ python main.py 

This starts the training for ResNet-20, saving the progress after training every 512 images. To train a net of different depth, comment the line in main.py

net = models.resnet(X, 20)

and uncomment the line initializing the appropriate model.