RT-BENE: A Dataset and Baselines for Real-Time Blink Estimation in Natural Environments

The RT-BENE dataset is licensed under CC BY-NC-SA 4.0. Commercial usage is not permitted. If you use our blink estimation code or dataset, please cite the relevant paper:

@inproceedings{CortaceroICCV2019W,
author={Kevin Cortacero and Tobias Fischer and Yiannis Demiris},
booktitle = {Proceedings of the IEEE International Conference on Computer Vision Workshops},
title = {RT-BENE: A Dataset and Baselines for Real-Time Blink Estimation in Natural Environments},
year = {2019},
}

More information can be found on the Personal Robotic Lab's website: https://www.imperial.ac.uk/personal-robotics/software/.

Overview

We manually annotated images that are contained in the "noglasses" part of the RT-GENE dataset with blink annotations. This dataset contains the extracted eye image patches and associated annotations.

In particular, rt_bene_subjects.csv is an overview CSV file with the following columns:

id

subject csv file

path to left eye images

path to right eye images

training/validation/discarded category

fold-id for the 3-fold evaluation.

Each individual "blink_labels" CSV file (s000_blink_labels.csv to s016_blink_labels.csv) contains two columns:

image file name

label, where 0.0 is the annotation for open eyes, 1.0 for blinks and 0.5 for annotator disagreement (these images are discarded)

Associated code

Please see the code repository for code allowing to train and evaluate a deep neural network based on the RT-BENE dataset. The code repository also links to pre-trained models and code for real-time inference.