Skip to content

Code for Zhejiang Lab Cup AI Competition (first round) [Multiple Object (Pedestrian) Tracking]

Notifications You must be signed in to change notification settings

YinqiangZhang/MOT_MUST1

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

24 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

之江杯(初赛)---行人多目标跟踪方法运行代码 MUST


依赖项

  • Cuda 8.0
  • Cudnn 7.0
  • Python 3.7
  • Pytorch 1.0.1
  • MATLAB R2017b

环境配置示例

conda create -n mot anaconda python=3.7
conda activate mot
conda install -c menpo opencv
conda install pytorch torchvision cudatoolkit=8.0 -c pytorch

ReID网络训练

(初赛使用,复赛已经更换)

  1. 数据集:Market1501(MIT license),相关说明在文件夹Market1501的readme.txt
  2. 数据集准备: 将数据集Market1501下载并保存至"./ReID/Market1501"到根目录中。运行prepare_market.py将数据集的结构修改成训练需要的形式(新的文件格式存在pytorch文件夹中,形成了训练用的train和val的文件夹,和测试用的query,gallery,multi-query的文件夹)
python prepare_market.py
  1. 训练:修改data_dir的路径,直接进行训练,可以修改参数,本次比赛使用参数在opts.yaml中。每次训练使用训练代码,数据处理文件,loss的变化情况以及当次实验用的参数也储存(备份)在"./ReID/model/ft_ResNet50"

  2. 验证: Test.py负责将测试数据(测试中只用query和gallery的数据进行验证)的特征提取出来并保存,方便后面评价方式rank@1,rank@5,rank@10,mAP的计算。 Evaluate.py负责计算上述评价方式。 (初赛的DMAN + ECO算法,参与视频b3的跟踪,使用的ReID方法与复赛相同,详情参见复赛说明)

使用方法

  1. 下载 ReID 已训练神经网络模型,保存到"./ReID/model/ft_ResNet50/"目录下
  2. 下载之江杯测试视频,保存到"./TrackingCode/data/MOT_ZJ/level1_vedio/"目录下
  3. 下载Yolov3权重文件,保存到"./YOLOv3/"文件夹
  4. 运行detect_video_YOLO.py生成标准格式的测试数据文件,位于 "./TrackingCode/data/MOT_ZJ/train/" 文件夹中
python detect_video_YOLO.py
  1. 进入 "./TrackingCode/ECO/" 文件夹中, 运行脚本 install.m 编译ECO跟踪器
  2. 运行服务器脚本
python similarity_calculate.py
  1. 在MATLAB中运行客户端脚本 MUST_demo.m 跟踪结果位于"./TrackingCode/results/"文件夹中
  2. 如获取比赛所需格式的txt结果文件,运行脚本文件,结果文件位于 "./TrackingCode/modified_results/"文件夹中
python resultFormatChange.py

References

[1] Danelljan, M., Bhat, G., Khan, F.S., Felsberg, M.: ECO: Efficient convolution operators for tracking. In: CVPR (2017)

[2] Xiang, Y., Alahi, A., Savarese, S.: Learning to track: Online multi-object tracking by decision making. In: ICCV (2015)

[3] Zhu, J., Yang, H., Liu, N., Kim, M., Zhang, W., Yang, M.: Online Multi-Object Tracking with Dual Matching Attention Networks. In: ECCV (2018)

[4] Redmon J , Farhadi A . YOLOv3: An Incremental Improvement. 2018.

[5] Zheng Z , Zheng L , Yang Y . A Discriminatively Learned CNN Embedding for Person Re-identification[J]. Acm Transactions on Multimedia Computing Communications & Applications, 2016, 14(1)

[6] Liang Zheng*, Shengjin Wang, Liyue Shen*, Lu Tian*, Jiahao Bu, and Qi Tian. Person Re-identification Meets Image Search. Technical Report, 2015. (*equal contribution)

About

Code for Zhejiang Lab Cup AI Competition (first round) [Multiple Object (Pedestrian) Tracking]

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •