Overview

This repository contains code for graph-level predictions on intersection data, represented in the form of graphs. The code expects your dataset to be formatted in a specific structure for proper processing.

Dependencies

Follow the below steps sequentially to download the code and install the dependencies-

git clone https://github.com/VTTI/GNN-based-intersection-safety.git

cd GNN-based-intersection-safety

docker build . -t gnn-safety

docker run -it --rm --runtime=nvidia -v {{dataPath}}:/data gnn-safety /bin/bash

Dataset Structure

Ensure your dataset follows the structure outlined below:

1. Individual folders named train and test, each containing a subfolder called raw.
2. Inside the raw folders, there should be:
   • Folder named jsongraphs containing JSON files representing graphs. (The bash script dot_to_json.sh can be used to convert DOT files to JSON)
   • Text file named jsonfiles.txt listing the names of each JSON file in the jsongraphs folder.
   • Labels file named labels.npy.

Code Files

dataset_featurizer.py: This file is responsible for extracting node features, edge features, and the adjacency matrix from the JSON graphs corresponding to each frame of the video.

model.py: This file contains details about the model architecture. Modify it according to your specific requirements. Note that not all models support both node and edge features. Refer to PyTorch Geometric GNN Cheat Sheet to choose an appropriate model based on your data characteristics.

train.py: Run this file once the dataset has been formatted and dependencies have been installed