Authors: Zeryab Moussaoui, Yacine Ben Ameur , Houssem Meghnouj
11 Mars 2019: Submit the code according to Alpha-Pilot deadline.
10 May 2019: Add technical report
The aim of this code is to complete the selection of the Lockheed Martin's Drone Race "Alpha Pilot" (https://www.herox.com/alphapilot/77-test-2) but can be used for another applications by using Transfer Learning.
The Alpha Pilot qualification evaluates team's skills in both Computer Vision and Machine Learning, by predicting gate locations on images :
DroGate is built on the efficient ResNet-based architecture. The convolution stage of the architecture consists of a fast ResNet-8 with 3 residual blocks, followed by dropout and ReLU non-linearity. The extracted features are then processed by two separates multilayer perceptrons to predict gate locations and their confidence score :
- Python (recommanded : 3.5)
- Tensorflow (recommanded : 1.12.0)
- OpenCV (recommanded : 4.0.0)
- Seaborn (recommanded : 0.9.0)
And also some librairies : numpy, python-wget, zipfile, json.
To (re) train your own Drogate, just follow instructions of Drogate_Training notebook.
Please read following article : DroGate, a Lightweight Real-Time Drone Perception System
If you use of our code or our report in your project , please cite :
@article{Drogate,
title={{Drogate}: a Lightweight Real-Time Drone Perception System},
author={Zeryab Moussaoui and al.},
journal={Github},
year={2019}
}