CheckDatabase

Visualize the Trainings-Data

Introduction

This chapter explains, how to check if the DeepDriving trainings database is installed correctly.

Pre-Requirements

• The caffe framework from this repository is build and installed correctly.

Download the Trainings Database

• You can download the original training database from the DeepDriving homepage: http://deepdriving.cs.princeton.edu/

  • It is called: TORCS_trainset (http://deepdriving.cs.princeton.edu/TORCS_trainset.zip)
  • Attention: The zip file is larger than 50GB!

• Unzip the training-data to a directory you want.

• (optional) Add the path to the training-data to the environment

export DEEPDRIVING_DATABASE_PATH=<path-to-database>

• (optional) Or add it permanently to all terminals:

echo "export DEEPDRIVING_DATABASE_PATH=<path-to-database>" >> ~/.bashrc
source ~/.bashrc

Start the Database Visualization Tool

• Run the visualization script from the caffe-repository:

cd <repository-path>/torcs
./torcs_visualize_database.sh

• You should see a window which shows a semantic situation description (lanes, position of host car, distance to traffic cars) and the corresponding frame from the database. On the terminal you can see the label-values for every frame.

• At the start, the application is in PAUSE mode. Focus the window and press p to disable the PAUSE mode.

• The following shortcuts are possible:

  • p: Enable or disable PAUSE mode.
  • n: Show next frame.
  • +: If the PAUSE mode is disabled, it increases the frame speed.
  • -: If the PAUSE mode is disabled, it decreases the frame speed.
  • ESC: Close application.

Visualize another Database

• If you want to visualize another database, just provide the path to this database as first argument to the script.

cd <repository-path>/torcs
./torcs_visualize_database.sh <path-to-alternative-database>

Next Step

• Drive in TORCS