Home

Awesome

Download and pre-process ROAD dataset

Here we provide the download and pre-processing instructions for the ROAD dataset, that is released through our TPAMI paper: ROAD: The ROad event Awareness Dataset for Autonomous Driving and uses 3D-RetinaNet code as a baseline, which also contains the evaluation code. The ROAD dataset will be used within The ROAD challenge.

Main Features

Attribution

ROAD dataset is build upon Oxford Robot Car Dataset (OxRD). If you find the original dataset useful in your work, please cite it using the citation that can be found here.

Similar to the original dataset (OxRD), the ROAD dataset is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License and is intended for non-commercial academic use. If you are interested in using the dataset for commercial purposes, please contact original creator OxRD for video content and Fabio and Gurkirt for event annotations.

If you use ROAD dataset, please cite it using the following:

@ARTICLE {singh2022road,
author = {Singh, Gurkirt and Akrigg, Stephen and Di Maio, Manuele and Fontana, Valentina and Alitappeh, Reza Javanmard and Saha, Suman and  Jeddisaravi, Kossar and Yousefi, Farzad and Culley, Jacob and Nicholson, Tom and others},
journal = {IEEE Transactions on Pattern Analysis & Machine Intelligence},
title = {ROAD: The ROad event Awareness Dataset for autonomous Driving},
year = {5555},
volume = {},
number = {01},
issn = {1939-3539},
pages = {1-1},
keywords = {roads;autonomous vehicles;task analysis;videos;benchmark testing;decision making;vehicle dynamics},
doi = {10.1109/TPAMI.2022.3150906},
publisher = {IEEE Computer Society},
address = {Los Alamitos, CA, USA},
month = {feb}
}

Download

BY DOWNLOADING THE DATASET VIDEOS YOU ARE BOUNDED TO ADHERE TO PRIVACY GUIDELINES OF OxRD. PLEASE VISIT OxRD PRIVACY POLICY FOR MORE DETAILS. THE VIDEOS FROM OxRD AND PROVIDED ANNOTATIONS ARE ONLY FOR ACADEMIC PURPOSE. 

We release the annotations created by Visual Artificial Intelligence Laboratory, and the sub-set of pre-processed videos from OxRD. Pre-processing includes demosaic for RGB conversion, ffmpeg for .mp4 conversion and fixing the frame-rate. More details can be found in tar2mp4.

You can download the Train-Val-set videos and corresponding annotations by changing your current directory to the road directory and running the bash file get_dataset.sh. This will automatically download the annotation files and video directory in the current directory (road).

bash get_dataset.sh

Alternatively, you can download the Train-Val-set videos and annotations from our Google-Drive folder.

The videos of Test-set and annotations are released, you can download it from our Google-Drive folder.

Frame-extraction

The baseline code for 3D-RetinaNet used in the dataset release paper uses sequences of frames as input. Once you have downloaded the videos from Google-Drive, create a folder name road and put the annotations under it, then create another folder named videos under road folder, and put all the videos under the folder named videos. Now, your folder structure should look like this:

    road/
        - road_trainval_v1.0.json
        - videos/
            - 2014-06-25-16-45-34_stereo_centre_02
            - 2014-06-26-09-53-12_stereo_centre_02
            - ........

Before extracting the frames, you will need to make sure that you have ffmpeg installed on your machine or your python should include its binaries. If you are using Ubuntu, the following command should be sufficient: sudo apt install ffmpeg.

You can now use extract_videos2jpgs.py to extract the frames. You will need to provide the path to the road folder as an argument:

python extract_videos2jpgs.py <path-to-road-folder>/road/

Now, the road directory should look like this:

    road/
        - road_trainval_v1.0.json
        - videos/
            - 2014-06-25-16-45-34_stereo_centre_02
            - 2014-06-26-09-53-12_stereo_centre_02
            - ........
        - rgb-images
            - 2014-06-25-16-45-34_stereo_centre_02/
                - 00001.jpg
                - 00002.jpg
                - .........*.jpg
            - 2014-06-26-09-53-12_stereo_centre_02
                - 00001.jpg
                - 00002.jpg
                - .........*.jpg
            - ......../
                - ........*.jpg

Annotation Structure

The annotations for the train and validation split are saved in single json file named road_trainval_v1.0.json, which is located under root directory of the dataset as it can be seen above.

The first level of road_trainval_v1.0.json contains dataset level information like classes of each label type:

Finally, the db field contains all frame and tube level annotations for all the videos:

Evaluation

Now that you have the dataset and are familiar with its structure, you are ready to train or test 3D-RetinaNet, which contains a dataloader class and evaluation scripts required for all the tasks in ROAD dataset.

You can find the evaluation functions in 3D-RetinaNet/modules/evaluation.py.

Plotting annotations

Please note that you need to setup the dataset structure like it is set in the Frame-extraction section! In order to inspect the dataset, you can use plot_annots.py to plot the annotations for the videos in road/rgb-images/. This will dump plotted images, then you can use ffmpeg to convert them into a video like shown below:

ROAD