- Implementation of NVIDIA's DAVE-2 neural network in pure PyTorch (PilotNet)
- Data augmentation in train-feed pipeline
- Online monitoring of the training process via Visdom
- VisualBackProp if it's of any help to sanity-checking the network
- Downloads all the data you need for your first run
- Easy-tweaking configurations
- Ready-to-use inference API
Simply run python3 main.py
to train a PilotNet from scratch. This should only take a few hours on a descent GPU.
This is a work of 3, done in the full-span of the 7th Hackathon of CafeBazaar -- two days! Results surprized the shit out of us!