Components Implemented :-
https://github.com/Sai-Venky/Trackjectory.git
.conda create --name track python=3.6
and activate it conda activate track
.pip install -r requirements.txt
.pip install lap cython-bbox
cd DCNV2
and running sh ./make.sh
.All configurable values with details on their significance are in utils/config.py
.
For testing, pre-trained models can be downloaded from
Siam RPN++
FairMOT DLA-34
Single Object Tracking :-
Training - python src/siam_train.py
Testing - python src/siam_track.py
Multi Object Tracking :-
Training - python src/mot_train.py
Testing - python src/mot_track.py
Trajectory Forcasting :-
Training - python src/trajectory_train.py
Testing - python src/trajectory_test.py
The dataset is created from Kabaddi player videos curated from multiple online platforms. (https://youtu.be/HOfY9g05Sv4) This was selected since this sport depicts a lot of movements (feints) and is challenging for forcasting trajectory correctly.
In order to train with a custom dataset, change the config value of multi_images_dataset
, single_track_dataset
, trajectory_dataset
in utils/config.py
.
The samples of how data should be constructed is provided in the data
folder.
The directory structure is as follows :-
trajectory_train : main file performing tasks for training Social GCN with outputs predicted from multi object tracker.
Thanks a lot for the wonderful work. The above pre-trained models are taken from the respective links.
https://github.com/ifzhang/FairMOT
https://github.com/zllrunning/SiameseX.PyTorch
https://github.com/abduallahmohamed/Social-STGCNN
You can contribute in serveral ways such as creating new features, improving documentation etc.
MIT Licence