anomaly_datasets

This repository can be used to generate the cumulated masks as in our paper "Perception Datasets for Anomaly Detection in Autonomous Driving: A Survey"

These provide an overview of how many anomalies are contained in a dataset and in which regions they can be found.

Structure

├── datasets                  # Configurations for individual datasets
├── figures                   # Output
  ├── overlays                # Images overlaid with ground truth
  └── pixel-distributions     # Accumulated anomaly masks    
└── helper                    # Helper functions for CODA, Vistas-NP, and WD-Pascal

Installation

It is recommended to create a python3 environment and install all required packages. Our code runs with python version 3.6

pip install --upgrade pip
pip install -r requirements.txt

Run Code

Before the code can be launched, the root paths to the datasets must be specified in config.yaml. Then you need to define for which dataset you want to create the cumulated masks. For this you need to change the dataset variable in config.yaml and run the following code:

python main.py

Hydra allows overwriting config values in the terminal. This can be done as below using the CWL dataset as an example:

python main.py ++dataset=cwl

Add a new dataset

If you want to add a dataset, you have to define your own class under the datasets folder. Feel free to use datasets/template.py for this purpose.

After that, another entry must be added to the configuration file according to the following scheme:

new_dataset:
  _target_: datasets.new_dataset.ND
  root: /PATH/TO/DATASET

Citation

If you find our work useful for your research, please cite our paper:

@InProceedings{Bogdoll_Perception_2023_IV,
  author    = {Bogdoll, Daniel and Uhlemeyer, Svenja and Kowol, Kamil and Z\"{o}llner, J. Marius},
  title     = {{Perception Datasets for Anomaly Detection in Autonomous Driving: A Survey}}, 
  booktitle = {Intelligent Vehicles Symposium (IV)},
  year      = {2023}
}