Udacity-SDC-Docker-GPU

• Docker files for Udacity SDC ND Term 1 and Term 3- GPU version

Term 1

Ubuntu 16.04
CUDA 8.0
cuDNN v6
python_version=3.5.2
miniconda_version=4.1.11
tensorflow_version=0.12.0rc0
Keras_version=1.1.2
iPython/Jupyter Notebook
Numpy, SciPy, Pandas, Scikit Learn, Matplotlib
opencv3, moviepy

Term 3

Ubuntu 16.04
CUDA 8.0
cuDNN v6
ROS kinetic
ROS kinetic perception
DataspeedInc/dbw_mkz_ros
Flask>=0.11.1
attrdict>=2.0.0
eventlet>=0.19.0
python-socketio>=1.6.1
numpy>=1.13.1
Pillow>=2.2.1
scipy
tensorflow-gpu (latest)
keras (latest)
h5py

Prerequisites

1. docker installation Install Docker following the installation guide for your platform.

   For details- https://docs.docker.com/engine/installation/

2. Nvidia-drivers Install supported nvidia drivers - 367 and above

3. CUDA-capable GPU Check if your system has CUDA-capable GPU

   Refer https://developer.nvidia.com/cuda-gpus

Build the image

Place the Dockerfile-<term> (rename to “Dockerfile”) in the current working directory and execute -
	$ docker build -t <dockerhub userid>/udacitysdc-<term>:gpu .

Once the image is built successfully, do a quick check with,
	$ docker images

You should find two images listed. 
	1. <dockerhub userid>/udacitysdc-<term>:gpu
	2. nvidia/cuda

Run the docker image in a container

We can now spin up one or more containers using this image.

$ nvidia-docker run -it -v $PWD:/src -p 8888:8888 <dockerhub userid>/udacitysdc-<term>:gpu bash

Note the use of nvidia-docker rather than just docker.

Sanity check

1. Check opencv3 for term 1. Run "Lane Finding Demo" as in the following link

https://medium.com/self-driving-cars/lane-finding-demo-bd834d7928a9#.i32ooenzi

Clone "lane-demo" package to your host folder ($PWD on host that maps to /src on Docker).
From the nvidia-docker terminal, run - 

	$ jupyter notebook --port=8888 --ip=0.0.0.0

From the browser, type-in "localhost:8888"

Run "Lane Finding Demo" and check if the demo runs successfully without errors.

2. Check tensorflow, gpu, keras From the nvidia-docker terminal, run -

   $ curl -sSL https://github.com/fchollet/keras/raw/master/examples/mnist_mlp.py | python

Verify that terminal prints messages for tensorflow and esblishing connections to GPU.

Check the output.

	60000 train samples
	10000 test samples
	Train on 60000 samples, validate on 10000 samples
	Epoch 0
	5s - loss: 0.4356 - acc: 0.8716 - val_loss: 0.1863 - val_acc: 0.9421
	.....
	.....
	.....
	Epoch 19
	5s - loss: 0.0383 - acc: 0.9874 - val_loss: 0.0705 - val_acc: 0.9820
	Test score: 0.0704572771238
	Test accuracy: 0.982

References:

1. https://docs.docker.com/engine/installation/
2. https://hub.docker.com/r/nvidia/cuda/
3. https://github.com/NVIDIA/nvidia-docker/wiki/CUDA#requirements
4. https://github.com/fchollet/keras/blob/master/docker/Dockerfile
5. http://ermaker.github.io/blog/2015/09/08/get-started-with-keras-for-beginners.html