DIVOTrack: A Novel Dataset and Baseline Method for Cross-View Multi-Object Tracking in DIVerse Open Scenes

This repository contains the details of the dataset and the Pytorch implementation of the Baseline Method CrossMOT of the Paper: DIVOTrack: A Novel Dataset and Baseline Method for Cross-View Multi-Object Tracking in DIVerse Open Scenes

Abstract

Cross-view multi-object tracking aims to link objects between frames and camera views with substantial overlaps. Although cross-view multi-object tracking has received increased attention in recent years, existing datasets still have several issues, including 1) missing real-world scenarios, 2) lacking diverse scenes, 3) owning a limited number of tracks, 4) comprising only static cameras, and 5) lacking standard benchmarks, which hinder the investigation and comparison of cross-view tracking methods. To solve the aforementioned issues, we introduce DIVOTrack: a new cross-view multi-object tracking dataset for DIVerse Open scenes with dense tracking pedestrians in realistic and non-experimental environments. Our DIVOTrack has ten distinct scenarios and 953 cross-view tracks, surpassing all cross-view multi-object tracking datasets currently available. Furthermore, we provide a novel baseline cross-view tracking method with a unified joint detection and cross-view tracking framework named CrossMOT, which learns object detection, single-view association, and cross-view matching with an all-in-one embedding model. Finally, we present a summary of current methodologies and a set of standard benchmarks with our DIVOTrack to provide a fair comparison and conduct a comprehensive analysis of current approaches and our proposed CrossMOT.

Dataset Description
- Dataset Structure
- Dataset Downloads
Training Detector
Single-view Tracking
Cross-view Tracking
Reference
Contact

The test result of the cross-view MOT baseline method MvMHAT on the DIVOTrack.

The ground truth of the DIVOTrack.

Dataset Description

We collect data in 10 different real-world scenarios, named: 'Circle', 'Shop', 'Moving', 'Park', 'Ground', 'Gate1', 'Floor', 'Side', 'Square', and 'Gate2'. All the sequences are captured by using 3 moving cameras: 'View1', 'View2', 'View3' and are manually synchronized.

In the old version, the corresponding scenarios named: 'circleRegion', 'innerShop', 'movingView', 'park', 'playground', 'shopFrontGate', 'shopSecondFloor', 'shopSideGate', 'shopSideSquare', 'southGate'. The corresponding camera is named: 'Drone', 'View1', 'View2'.

For the test set, we provide the ground truth of the 5 scenes: 'Circle', 'Gate1', 'Floor', 'Shop', and 'Square'.

Dataset Structure

The structure of our dataset is as follows:

DIVOTrack
    └─────datasets
             └─────DIVO
                    ├───images
                    │    ├───annotations
                    │    ├───dets
                    │    ├───train
                    │    └───test
                    ├───labels_with_ids
                    │    ├───train
                    │    └───test
                    ├───ReID_format
                    │    ├───bounding_box_test
                    │    ├───bounding_box_train
                    │    └───query
                    └───boxes.json

Dataset Downloads

The whole dataset can be downloaded from Huggingface. Note that, each file needs to unzip by the password. You can decompress each .zip file in its folder after sending us (shengyuhao@zju.edu.cn, gaoangwang@intl.zju.edu.cn) the License in any format. After that, you should run generate_ini.py to generate seqinfo.ini file.

Training Detector

The training process of our detector is in ./Training_detector/ and the details can be seen from Training_detector/README.md.

Single-view Tracking

We conducted experiments on DIVOTrack in five benchmarks:

Benchmark	HOTA ↑	IDF1 ↑	MOTA ↑	MOTP ↑	MT ↑	ML ↓	AssA ↑	IDSw ↓	FM ↓
DeepSort	54.3	59.9	79.6	81.2	462	50	45.0	1,920	2,504
CenterTrack	55.3	62.2	73.4	80.6	534	35	49.2	1,631	2,950
Tracktor	48.4	56.2	66.6	80.8	517	22	40.3	1,382	3,337
FairMOT	65.3	78.2	82.7	81.9	486	48	62.7	731	3,498
TraDeS	58.9	67.3	74.2	82.3	504	38	54.0	1,263	2,647

Each single-view tracking baseline is evaluated using TrackEval.

Cross-view Tracking

We conducted experiments on the DIVOTrack dataset using six benchmarks as well as our proposed method CrossMOT	Benchmark	CVMA ↑
OSNet	34.3	46.0
Strong	40.9	45.9
AGW	57.0	56.8
MvMHAT	61.0	62.6
CT	64.9	65.0
MGN	33.5	39.4
CrossMOT	72.4	71.1

With the exception of CrossMOT, all of the other Re-ID methods require Multi_view_Tracking to predict the tracking results after they are obtained. Finally, the results of CVMA and CVIDF1 are obtained through MOTChallengeEvalKit_cv_test.

Reference

Any use whatsoever of this dataset and its associated software shall constitute your acceptance of the terms of this agreement. By using the dataset and its associated software, you agree to cite the papers of the authors, in any of your publications by you and your collaborators that make any use of the dataset, in the following format:

@article{hao2024divotrack,
  title={Divotrack: A novel dataset and baseline method for cross-view multi-object tracking in diverse open scenes},
  author={Hao, Shengyu and Liu, Peiyuan and Zhan, Yibing and Jin, Kaixun and Liu, Zuozhu and Song, Mingli and Hwang, Jenq-Neng and Wang, Gaoang},
  journal={International Journal of Computer Vision},
  volume={132},
  number={4},
  pages={1075--1090},
  year={2024},
  publisher={Springer}
}

The license agreement for data usage implies the citation of the paper above. Please notice that citing the dataset URL instead of the publications would not be compliant with this license agreement. You can read the LICENSE from LICENSE.

Contact

If you have any concerns, please contact shengyuhao@zju.edu.cn

shengyuhao / DIVOTrack

readme