This is my take on the speed challenge proposed by comma.ai. You can check out the link to the original contenst page here https://github.com/commaai/speedchallenge.
Download the dataset from the commaai repo and extract it.
Prepare the data set by running ./preprocessor.py - This will save the images in a folder called data_preprocessed as well as a file called preprocessed.csv including path to image, time stamp and speed. This file is used by both the training script and the predict script.
Train the model by running ./training.by - This will save the weights in the file model-weights.h5.
Predict by running ./predict.py - This will use the saved model weights and test on some test data from preprocessed.py.