Group-Activity-Recognition
Group-Activity-Recognition copied to clipboard
A novel Participation-Contributed Temporal Dynamic Model for Group Activity Recognition
Participation-Contributed Temporal Dynamic Model for Group Activity Recognition. PDF
We give a general DMS(Data, Model, Solver) code framework for PCTDM, impelemented by Pytorch. You can apply new model or dataset into this framework by modifying the files in Configs
easily! For further information about me, welcome to my homepage.
Requirements
> Ubuntu 16.04
> pytorch 0.4.1
= python 2.7
pip install dlib
The general piplines of GAR
You can run python GAR.py
to excute all the following steps.
Step Zero: Preprocessing dataset
- To download VD and CAD at './dataset/VD' and './dataset/CAD' folder;
- Add
none.jpg
- To track the persons and generate the train/test files by using Processing.py;
Step One: Action Level
- To create a
Piplines
instance as:
Action = Action_Level(dataset_root, dataset_name, 'trainval_action')
;
- For action recognition, you can use
Action.trainval()
; - For extracting action features, you can use
Action.extract_feas(save_folder='*')
.
Step Two: Activity Level
This is the core part of GAR which need to be designed by youself. We proposed a novel PCTDM to aggreate the action features with attending to key persons.
- To create a
Piplines
instance as:
Activity = Activity_Level(dataset_root, dataset_name, 'trainval_activity')
;
- For activity recognition, you can use
Activity.trainval()
.
Step Three: Evaluate
- To show some results at activity level, you can use
Activity.evaluate()
.
All steps may take about 15 hours for 'VD', and 5 hours for 'CAD'.
License and Citation
Please cite the following paper in your publications if it helps your research.
@inproceedings{yan2018participation,
title={Participation-Contributed Temporal Dynamic Model for Group Activity Recognition},
author={Yan, Rui and Tang, Jinhui and Shu, Xiangbo and Li, Zechao and Tian, Qi},
booktitle={2018 ACM Multimedia Conference on Multimedia Conference},
pages={1292--1300},
year={2018},
organization={ACM}
}
Contact Information
Feel free to create a pull request or contact me by Email = ["ruiyan", at, "njust", dot, "edu", dot, "cn"], if you find any bugs.