Final term project in the course TDT4265 - Computer vision and deep learning of spring 2019 at NTNU
Implemented by Sondre Aleksander Bergum, Martin Madsen and Filip Schjerven.
Train a model by running model.py. By default the code saves it as "model.h5".
To change what training data your model trains on you must manually change to the correct .csv and image folder in main, the default is to train a model on data from both tracks.
We construct visualizations of layer activations from a random training-image after training a model that are saved in docs/plots.
Run models by executing "python3 drive.py " in terminal. Some good models are provided for you already.
Project description (Project 3)
Provided code
Litterature:
https://arxiv.org/pdf/1704.07911.pdf
https://arxiv.org/pdf/1604.07316.pdf
https://arxiv.org/pdf/1710.03804.pdf
https://selfdrivingcars.mit.edu/ <MIT 6.S094: Deep Learning for Self-Driving Cars>
http://cs231n.stanford.edu/reports/2017/pdfs/626.pdf
https://arxiv.org/pdf/1608.01230.pdf + https://github.com/commaai/research
https://devblogs.nvidia.com/explaining-deep-learning-self-driving-car/
https://devblogs.nvidia.com/deep-learning-self-driving-cars/
Other resources:
https://blog.coast.ai/training-a-deep-learning-model-to-steer-a-car-in-99-lines-of-code-ba94e0456e6a
https://github.com/tech-rules/DAVE2-Keras
https://github.com/pszczesnowicz/SDC-P3-Behavioral-Cloning/blob/master/model.py
GPLv2