Prediction Models for Drivable Area

Description

This repository contains the deep learning models and techniques used to predict the changes in the drivable area of an ego vehicle. The input is an array of 100 y-coordinates representing the boundary of the drivable area polygon. The output is an array representing the predicted drivable area for the recent future.

How to Run

1. First, install the dependencies.

   pip install -r requirements.txt
   

2. Create a .env file using the env_template.txt file.

   You can set N_TH_FRAME = True to predict only for the nth frame in the future and not for all n frames in the future. Set DATASET_TYPE = "simple" to generate simple drivable areas with consistent velocity, and set DATASET_TYPE = "video" to run for other drivable area datasets. Numpy files corresponding to other datasets can be found here

3. To run for other datasets, download the dataset from the above link and copy the local folder path to the directory_path variable under the VideoFrameDataset initiation in the run.py file.

4. Run the module

   python3 run.py
   

To Run a Collision Classifier

Running a pretrained Collision Classifier

1. Modify the following environment variables -

COLLISION_FLAG = True
PRETRAINED_FLAG = True
DATASET_TYPE = "collision"
DATASET_PATH = "path/to/collision/data"
COLLISION_MODEL_NAME = "classifier_model_name"

2. Run the module

   python3 run.py