Models.md

TrailNet models

This document describes how to create TrailNet model from scratch using NVIDIA DIGITS.

Creating a dataset

First, we need to create a dataset in DIGITS. Follow the steps in this document.

Using an existing dataset

The IDISA Forest Trail dataset is publicly available here. Follow the steps in this document.

For our dataset (Zografou University Campus) follow the steps in this document.

Training a model

Once the dataset is created, the model can be trained and fine-tuned using DIGITS. Follow the steps in this document.

Deploying a model

Once the model is trained, resulting .prototxt and .caffemodel files can be used by ROS nodes running on Jetson.

Pre-trained models

You can also use pre-trained models as described in here. AriadneNet is our trained model based on the TrailNet architecture.

Network Architectures (TrailNet)

The architecture used for our trail following DNN, the AriadneNet, is based on TrailNet. More information can be found here.