/ViT-DD

This repo contains code and models for the implementation of ViT-DD, a semi-supervised method for detecting driver distractions.

Primary LanguagePythonMIT LicenseMIT

ViT-DD: Multi-Task Vision Transformer for Semi-Supervised Driver Distraction Detection

paper

ViT-DD

Abstract

Driver distraction detection is an important computer vision problem that can play a crucial role in enhancing traffic safety and reducing traffic accidents. This paper proposes a novel semi-supervised method for detecting driver distractions based on Vision Transformer (ViT). Specifically, a multi-modal Vision Transformer (ViT-DD) is developed that makes use of inductive information contained in training signals of distraction detection as well as driver emotion recognition. Further, a self-learning algorithm is designed to include driver data without emotion labels into the multi-task training of ViT-DD. Extensive experiments conducted on the SFDDD and AUCDD datasets demonstrate that the proposed ViT-DD outperforms the best state-of-the-art approaches for driver distraction detection by 6.5% and 0.9%, respectively.

Results

Experiments Accuracy NLL Checkpoints
AUCDD 0.9359 0.2399 link
SFDDD split-by-driver 0.9251 0.3900 link
SFDDD split-by-image 0.9963 0.0171 link

Usage

Prerequisites

The code is built with following libraries:

Data Preparation

Please organize the data using the directory structures listed below:

data_root
|-- AUCDD
    |-- v2
        |-- cam1
            |-- test
            |-- train
              |-- c0
              |-- ...
              |-- c9
                |-- 188.jpg
                |-- ...
|-- SFDDD 
    |-- imgs
        |-- train
          |-- c0
          |-- ...
          |-- c9
            |-- img_19.jpg
            |-- ...

pseudo_label_path
|-- AUCDD
  |-- emo_list.csv
  |-- imgs
      |-- c0
      |-- ...
      |-- c9
          |-- 0_face.jpg
          |-- ...
|-- SFDDD
  |-- emo_list.csv
  |-- imgs
      |-- img_5_face.jpg
      |-- ...

We provide our generated pseudo emotion labels as well as cropped images of drivers' faces for the AUCDD and SFDDD datasets here.

Citation

If you find ViT-DD beneficial or relevant to your research, please kindly recognize our efforts by citing our paper:

@article{Ma2022MultiTaskVT,
  title={Multi-Task Vision Transformer for Semi-Supervised Driver Distraction Detection},
  author={Yunsheng Ma and Ziran Wang},
  journal={arXiv},
  year={2022}
}