Home

Welcome to the DeepLearningCaroloCup wiki!

The purpose of this repository is to archive the work form the 2016 CaroloCup for future teams.

Dependencies:

• Tensorflow version 1.3
• Python 2.7 or 3.5.4

General Steps

Once you have a tensorflow helloworld! example running, clone the repo.

• Clone Repo

Next, downloaded the dataset.

• Download dataset: https://github.com/udacity/self-driving-car/tree/master/annotations Download Data Set 1.

Preparing the dataset before training steps:

• Set FILEPATH_IMAGE and FILEPATH_IMAGE in createRecord.py
• Run createRecord.py - creates a new .tfrecord file in dir TFRecord. A TFRecord is a binary file with a set structure.

Training steps:

• Run train.py - will read the tfrecord in batches.

Notes:

• Training logs and saved model weights will be created in directory: tf_logs
• Accuracy of model of the test set is displayed on the terminal at the end of the complete training process.
• Saved Model Accuracy: 72%.