This is the pytorch implementatin of Semantics-guided Part Attention Network (SPAN)
Orientation-aware Vehicle Re-identification with Semantics-guided Part Attention Network
Tsai-Shien Chen, Chih-Ting Liu, Chih-Wei Wu, and Shao-Yi Chien
European Conference on Computer Vision (ECCV), Oral, 2020
[Paper Website] [arXiv]
If you use SPAN, please cite this paper:
@inproceedings{SPAN,
title = {Orientation-aware Vehicle Re-identification with Semantics-guided Part Attention Network},
author = {Chen, Tsai-Shien and Liu, Chih-Ting and Wu, Chih-Wei and Chien, Shao-Yi},
booktitle = {European Conference on Computer Vision},
pages = {330--346},
year = {2020},
organization = {Springer}
}
We visiualize some examples of vehicle images and their
$ git clone https://github.com/tsaishien-chen/SPAN.git
$ cd SPAN
requirements.txt
. You can install all the dependancies by:
$ sed -i 's/==/>=/g' requirements.txt
$ pip3 install -r requirements.txt
To train the model from scratch, please run
$ python3 main.py --mode train --image_root <Path_to_VeRi>
For example,
$ python3 main.py --mode train --image_root ../Dataset/VeRi
The whole training process includes five steps:
We have given the pretrained model of part attention generator; therefore, you can simply generate the part attention mask without training by
$ python3 main.py --mode implement --image_root <Path_to_VeRi>
For example,
$ python3 main.py --mode implement --image_root ../Dataset/VeRi
After training and implementation process, the code will automatically visualize generated masks as above.
Or, you can uncomment the visualize
function in main.py
and can independently visualize the masks after being generated in each step.
Tsai-Shien Chen, Media IC and System Lab, National Taiwan University E-mail : tschen@media.ee.ntu.edu.tw