This repository includes the code and dataset information required for reproducing the results in our paper. Besides, we also integrated the source code of our baseline method, DeepTL-Lane-Change-Classification, into this repo. The baseline approach infers the risk level of lane change video clips with deep CNN+LSTM.
As for fabricating the lane-changing datasets, we use Carla CARLA 0.9.8 which is an open-source autonomous car driving simulator. Besides, we also utilized the scenario_runner which was designed for CARLA challenge event. For real-driving datasets, we used Honda-Driving Dataset (HDD) in our experiments. We published the converted scene-graph datasets used in our paper here. **brandon fix this.
The architecture of this repository is as below:
- sg-risk-assessment/: this folder collects all the core model/trainer/utilties used for our scene-graph based approach.
- baseline-risk-assessment/: this folder collects all the related source file that our baseline method requires
- Mask_RCNN is the module that handle object detection and coloring on the image sequence.
- sg_risk_assessment.py: the script that triggers our scene-graph based approach.
- baseline_risk_assessment.py: the script that triggers our baseline algorithm.
We recommend our potential users to use Anaconda as the primary virtual environment.
$conda install python=3.6.8
$conda install -c anaconda cython=0.29.10
$conda install -c aaronzs tensorflow-gpu
$conda install git
$pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI
$pip install -r requirements.txt
Since our primary working environment is Windows, we refer to this solution to have pycocotools to be installed.
To be filled.
-
how to run sg_risk_assessment.py
-
how to run baseline_risk_assessment.py
-
what are the expected outputs?
-
Information about pretrained coco model: Mask-RCNN pretrained weights can refer to this link and move it under [project folder]/pretrained_model (credit to [this repo](Mask R-CNN implementation by Matterport).
Please kindly consider citing our paper if you find our work useful for your research
@article{yu2020scene,
title={Scene-graph augmented data-driven risk assessment of autonomous vehicle decisions},
author={Yu, Shih-Yuan and Malawade, Arnav V and Muthirayan, Deepan and Khargonekar, Pramod P and Faruque, Mohammad A Al},
journal={arXiv preprint arXiv:2009.06435},
year={2020}
}