Docker Environment

Features

• Three docker images: Tensorflow, Keras, Caffe
• Tensorflow object detection API
• NVIDIA GPU
• Display webcam video in docker container via opencv windows

Installation

Ubuntu 16.04

For developer environments

$ sudo ./install_docker_ce_ubuntu_amd64.sh  # Install docker-ce
$ sudo ./install_nvidia_docker_ubuntu_amd64.sh  # Install nvidia-docker
$ sudo mkdir /root/Data  # Our convention is to put training data here. This folder will be mounted to the docker container.

Create environment in Docker

Tensorflow

To build the clean Tensorflow 1.8 (GPU) environment

$ cd {PROJECT_ROOT}/framework/tensorflow/gpu
$ sudo make build   # Build tensorflow DockerFile

If you want to build the clean Tensorflow 1.8 (GPU) environment

$ sudo make build TF_VER=1.8.0

To run the clean Tensorflow (GPU) environment

$ cd {PROJECT_ROOT}/framework/tensorflow/gpu
$ sudo make x11 # or make x11 TF_VER=1.8.0

Keras with Tensorflow backend

To build the clean Keras (GPU) environment

$ cd {PROJECT_ROOT}/framework/keras/gpu
$ sudo make build   # Build keras docker file

To run the clean Keras (GPU) environment

$ cd {PROJECT_ROOT}/framework/keras/gpu
$ sudo make x11

Run the clean Keras (GPU) environment on the web (IPython notebook and TensorBoard)

$ cd {PROJECT_ROOT}/framework/keras/gpu
$ sudo make notebook

You will see the notebook and Tensorboard URL in the terminal

Caffe

To run the clean caffe environment

$ cd framework/caffe
$ sudo make bash

To Test is caffe environment correct. Let's train LeNet on MNIST

$ cd $CAFFE_ROOT
$ ./data/mnist/get_mnist.sh
$ ./examples/mnist/create_mnist.sh
$ ./examples/mnist/train_lenet.sh

If you see "Optimization Done" in the console like following. Congratulation!

I1129 03:51:48.452687    39 caffe.cpp:259] Optimization Done.