Official implementation in PyTorch of Chained-Tracker as described in Chained-Tracker: Chaining Paired Attentive Regression Results for End-to-End Joint Multiple-Object Detection and Tracking.
The introduction video of CTracker is uploaded to Youtube.
The codes is tested with PyTorch 0.4.1. It may not run with other versions.
- Clone this repo into a directory named CTRACKER_ROOT
- Install the required packages
apt-get install tk-dev python-tk
- Install Python dependencies. We use python 3.6.5 and pytorch 0.4.1
conda create -n CTracker
conda activate CTracker
conda install pytorch=0.4.1 cuda90 -c pytorch
cd ${CTRACKER_ROOT}
pip install -r requirements.txt
sh lib/build.sh
MOT17 dataset can be downloaded at MOTChallenge.
We uses two CSV files to organize the MOT17 dataset: one file containing annotations and one file containing a class name to ID mapping.
We provide the two CSV files for MOT17 with codes in the CTRACKER_ROOT/data, you should copy them to MOT17_ROOT before starting training.
MOT17_ROOT/
|->train/
| |->MOT17-02/
| |->MOT17-04/
| |->...
|->test/
| |->MOT17-01/
| |->MOT17-03/
| |->...
|->train_annots.csv
|->train_labels.csv
MOT17_ROOT is your path of the MOT17 Dataset.
The CSV file with annotations should contain one annotation per line. Images with multiple bounding boxes should use one row per bounding box. Note that indexing for pixel values starts at 0. The expected format of each line is:
path/to/image.jpg,id,x1,y1,x2,y2,class_name
The MOT17 CSV file can be generated by generate_csv.py:
python generate_csv.py
You can modify this script to handle other datasets.
The class name to ID mapping file should contain one mapping per line. Each line should use the following format:
class_name,id
Indexing for classes starts at 0. Do not include a background class as it is implicit.
For example:
person,0
The network can be trained using the train.py script. For training on MOT17, use
CUDA_VISIBLE_DEVICES=0 python train.py --root_path MOT17_ROOT --model_dir ./ctracker/ --depth 50
By default, testing will start immediately after training finished.
A trained model is available at Google Drive/Tencent Weiyun, run the following commands to start testing:
CUDA_VISIBLE_DEVICES=0 python test.py --dataset_path MOT17_ROOT --model_dir ./trained_model/
- Part of codes are borrowed from the pytorch retinanet implementation
- The NMS module used is from the simpledet
If you find CTracker is useful in your project, please consider citing us:
@inproceedings{peng2020ctracker,
title={Chained-Tracker: Chaining Paired Attentive Regression Results for End-to-End Joint Multiple-Object Detection and Tracking},
author={Peng, Jinlong and Wang, Changan and Wan, Fangbin and Wu, Yang and Wang, Yabiao and Tai, Ying and Wang, Chengjie and Li, Jilin and Huang, Feiyue and Fu, Yanwei},
booktitle={Proceedings of the European Conference on Computer Vision},
year={2020},
}