NII Face Mask Dataset

=====================================================================<br> # NII Face Mask Dataset v1.0<br> ===================================================================== Authors:<br> Trung-Nghia Le (1), Khanh-Duy Nguyen (2), Huy H. Nguyen (1), Junichi Yamagishi (1), Isao Echizen (1) Affiliations:<br> (1)National Institute of Informatics, Japan <br> (2)University of Information Technology-VNUHCM, Vietnam National Institute of Informatics <br> Copyright (c) 2021 Emails:<br> {ltnghia, nhhuy, jyamagis, iechizen}@nii.ac.jp, {khanhd}@uit.edu.vn Arxiv: https://arxiv.org/abs/2111.12888<br> NII Face Mask Dataset v1.0: https://zenodo.org/record/5761725 =============================== INTRODUCTION =============================== The NII Face Mask Dataset is the first large-scale dataset targeting mask-wearing ratio estimation in street cameras. This dataset contains 581,108 face annotations extracted from 18,088 video frames (1920x1080 pixels) in 17 street-view videos obtained from the Rambalac's YouTube channel. - https://www.youtube.com/c/Rambalac The videos were taken in multiple places, at various times, before and during the COVID-19 pandemic. The total length of the videos is approximately 56 hours. <br> =============================== REFERENCES =============================== If your publish using any of the data in this dataset please cite the following papers: #Pre-print version<br> @article{Nguyen202112888,<br> title={Effectiveness of Detection-based and Regression-based Approaches for Estimating Mask-Wearing Ratio},<br> author={Nguyen, Khanh-Duy and Nguyen, Huy H and Le, Trung-Nghia and Yamagishi, Junichi and Echizen, Isao},<br> archivePrefix={arXiv},<br> arxivId={2111.12888},<br> url={https://arxiv.org/abs/2111.12888},<br> year={2021}<br> } #Final version<br> @INPROCEEDINGS{Nguyen2021EstMaskWearing,<br> author={Nguyen, Khanh-Duv and Nguyen, Huv H. and Le, Trung-Nghia and Yamagishi, Junichi and Echizen, Isao},<br> booktitle={2021 16th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2021)}, <br> title={Effectiveness of Detection-based and Regression-based Approaches for Estimating Mask-Wearing Ratio}, <br> year={2021},<br> pages={1-8},<br> url={https://ieeexplore.ieee.org/document/9667046},<br> doi={10.1109/FG52635.2021.9667046}} <br> ======================== DATA STRUCTURE ================================== <br> 1. Directory Structure<br> ------------------------------- ./NFM<br> ├── dataset<br> │ ├── train.csv: annotations for the train set.<br> │ ├── test.csv: annotations for the test set.<br> └── README_v1.0.md <br> 2. Description for each files in detail.<br> --------------------------------------------------------- We use the same structure for two CSV files (train.csv and test.csv). Both CSV files have the same columns:<br> &lt;1st column&gt;: video_id (a source video can be found by following the link: https://www.youtube.com/watch?v=&lt;video_id&gt;)<br> &lt;2nd column&gt;: frame_id (the index of a frame extracted from the source video)<br> &lt;3rd column&gt;: timestamp in milisecond (the timestamp of a frame extracted from the source video)<br> &lt;4th column&gt;: label (for each annotated face, one of three labels was attached with a bounding box: 'Mask'/'No-Mask'/'Unknown')<br> &lt;5th column&gt;: left<br> &lt;6th column&gt;: top<br> &lt;7th column&gt;: right<br> &lt;8th column&gt;: bottom<br> Four coordinates (left, top, right, bottom) were used to denote a face's bounding box. <br> ============================== COPYING ================================ This repository is made available under Creative Commons Attribution License (CC-BY). Regarding Creative Commons License: Attribution 4.0 International (CC BY 4.0), <br> please see https://creativecommons.org/licenses/by/4.0/ THIS DATABASE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND <br> ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED <br> WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. <br> IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, <br> INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, <br> BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, <br> OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, <br> WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) <br> ARISING IN ANY WAY OUT OF THE USE OF THIS DATABASE, EVEN IF ADVISED OF THE <br> POSSIBILITY OF SUCH DAMAGE <br> ====================== ACKNOWLEDGEMENTS ================================ This research was partly supported by JSPS KAKENHI Grants (JP16H06302, JP18H04120, JP21H04907, JP20K23355, JP21K18023), and JST CREST Grants (JPMJCR20D3, JPMJCR18A6), Japan. This dataset is based on the Rambalac's YouTube channel: https://www.youtube.com/c/Rambalac<br>