Keras implementation of End to End Learning for Self-Driving Cars by Nvidia.
Many thanks to Sully Chen for the original TensorFlow implementation and for sharing his own dataset, used here for training.
The network consists of 5 convolution layers (three 5x5 and two 3x3), and 4 fully connected layers on top.
To start training run python main.py. The data needs to be in the driving_dataset folder.