Home

Welcome to the Vehicle_Collision_Prediction_Using_CNN-LSTMs wiki! Here we can how to run the file in more details.

Data Collection

Start Carla with low resolution.
You may have to checkout the Python API folder to see if you can customize your script.
Run the script Autopilot script to collect and store the autopilot data.
Do the same using Random collision maker script and now shuffle and convert this data into numpy format with batches of 8.
make train, test and valid splits and put it into files folder in the root directory.

Nothing is needed to be changed to run the program as it is.
Default mode is testing. change the mode in config.py to "train" and mention the model type as "inception" or "vgg"
if you want to modify the network then exercise caution. However, you can play around with the hyperparameters to get better results.
Open the config.py to see the available list of options. You can see the hyperparameters and folders to store all the data.
run avntk.py to start the training.

To test it on a random video, use the same script but switch to "test" mode in the configuration.
mention the input and output filenames in the test file.
Run the program avntk.py and the output video will be generated in the mentioned location.