By Fu-Hsiang Chan, Yu-Ting Chen, Yu Xiang, Min Sun.
Anticipating Accidents in Dashcam Videos is initially described in a ACCV 2016 paper. We propose a Dynamic-Spatial-Attention (DSA) Recurrent Neural Network (RNN) for anticipating accidents in dashcam videos.
-
Dataset : link (Download the file and put it in "datatset/videos" folder.)
-
CNN features : link (Download the file and put it in "dataset/features" folder.)
-
Annotation : link
If you need the ground truth of object bounding box and accident location, you can download it.
The format of annotation:
<image name, track_ID, class , x1, y1, x2, y2, 0/1 (no accident/ has accident)>
python accident.py --model ./demo_model/demo_model
python accident.py --mode train --gpu gpu_id
python accident.py --mode test --model model_path --gpu gpu_id
Please cite this paper in your publications if you use this code for your research:
@inproceedings{chan2016anticipating,
title={Anticipating accidents in dashcam videos},
author={Chan, Fu-Hsiang and Chen, Yu-Ting and Xiang, Yu and Sun, Min},
booktitle={Asian Conference on Computer Vision},
pages={136--153},
year={2016},
organization={Springer}
}