Behaviorial Cloning Project

Overview

This repository contains a few models to tackle Udacity's Behavioral Cloning Project.

Besides the seminal PilotNet from NVIDIA, a couple of models pretrained on imagenet are included as well: MoblineNet and VGG16.

All these models are implemented using Keras.

The first step is downloading Udacity' Simulator where training examples are generated and the models validated. Click here for the details.

The included saved nvidia model and example videos were trained on the dataset provided by Udacity which can be downloaded from the classroom.

Training any model requires that a training dataset is present in ./data:

$ python train.py {nvidia,mobilenet,vgg16}

Validation on the simulator is accomplished by using the drive.py script as follows:

$ python drive.py [model.h5]

If using pip the dependencies can be installed by

$ pip install -r requirements.txt

This project requires: