Netspresso 团队 (Nota Inc.) 第七届 NVIDIA AI 城市挑战赛(赛道 1:多摄像头人物追踪)的官方资料库
bash ./setup.sh
docker build -t aic2023/track1_nota:latest -f ./Dockerfile .
docker run -it --gpus all -v /path/to/AIC2023_Track1_Nota:/workspace/AIC2023_Track1_Nota aic2023/track1_nota:latest /bin/bash
# extract frames
python3 tools/extract_frames.py --path /path/to/AIC23_Track1_MTMC_Tracking/
确保数据结构如下:
├── AIC2023_Track1_Nota
└── datasets
| ├── S001
| | ├── c001
| | | ├── frame1.jpg
| | | └── ...
| | ├── ...
| | └── map.png
| ├── ...
| └── S022
|
└── pretrained
├── market_mgn_R50-ibn.pth
├── duke_sbs_R101-ibn.pth
├── msmt_agw_S50.pth
├── market_aic_bot_R50.pth
├── yolov8x6.pth
├── yolov8x6_aic.pth
└── yolov8x_aic.pth
运行bash ./run_mcpt.sh
结果文件将保存如下:
├── AIC2023_Track1_Nota
└── results
├── S001.txt
├── ...
└── track1_submission.txt
@InProceedings{Kim_2023_CVPR,
author = {Jeongho Kim, Wooksu Shin, Hancheol Park and Jongwon Baek},
title = {Addressing the Occlusion Problem in Multi-Camera People Tracking with Human Pose Estimation},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2023},
}
该存储库中发布的多摄像头人员跟踪系统是通过组合多个模块(例如对象检测器、重新识别模型、多对象跟踪模型)而开发的。不允许将任何修改、添加或新训练的参数用于组合这些模块的商业用途。然而,未经修改的模块的商业使用是允许在其各自的许可证下进行的。如果您希望将各个模块用于商业用途,您可以参考下面提供的原始存储库和许可证。
对象检测器(许可证)链接:Github、许可证
重新识别模型(许可证)链接:Github、License
多目标跟踪模型(许可证)链接:Github、License