Lab: Getting Started

Udacity GPU Jupyter Notebook Workspace

Udacity provides an in-classroom Workspace with Jupyter Notebook and GPU support. Now you could complete the lab in the classroom! It is in the last concept of this lesson: Jupyter Workspace - Semantic Segmentation. Open it in another tab to work on it along the lesson.

If you are not familiar with it, check out the introductions in the project lesson here: Udacity GPU Workspace Introduction

Note that if you decided to use the Workspace, you could safely skip the local environment setup below!

Local Lab Setup

Make sure you have followed the instructions in the classroom to setup your environment or have followed along in the previous lab notebook setups. Please note that dependencies may be missing with a old setup. Check here for package requirements and manual installation of them.

Clone the Repository and Run the Notebook

Run the commands below to clone the lab repository and then run the notebook:

git clone https://github.com/udacity/RoboND-Segmentation-Lab.git
# Make sure your conda environment is activated!
jupyter notebook

The Jupyter interface will open in your browser. You can then access the cloned repo and the Jupyter Notebook from there. We are specifically working with the segmentation_lab.ipynb which can be found in following path code/segmentation_lab.ipynb.

Download the Data

After you have the notebook up and running be sure to download the training and validation data. Then put the respective folders in the /data directory.

Once the notebook is up and running and the data is downloaded, you can follow the instructions in the notebook and fill out the required pieces of code marked by TODOs. It is important to take time and read the comments in the notebook. On top of following along with the classroom lessons for guidance on how to fill out the TODOs be sure to check the notebook for relevant information as well. By the end you will have your first basic implementation of the network needed to get the project running!

It is important to note that some computer platforms may take up to 3 hours to train the network, depending on a few factors.

The recommended strategy for dealing with this problem is to complete the coding and debugging on your local system before moving to a faster system for the training portion. Once your network is running correctly you can then launch your notebook from your AWS instance in order to speed up training times. More information on running a Jupyter Notebook from AWS can be found here.