TrackNetV2-pytorch

Paper: TrackNetV2: Efficient Shuttlecock Tracking Network

Original Project（tensorflow）: https://nol.cs.nctu.edu.tw:234/open-source/TrackNetv2

官方上传的标注工具、数据集均已失效。del>

The author has now reuploaded the dataset。

Paper reading：TrackNetV2论文记录与pytorch复现

Inference with pytorch weights converted from tensorflow weights:

git apply tf2torch/diff.txt
python detect.py --source xxx.mp4 --weights  ./tf2torch/track.pt --view-img		# TrackNetv2/3_in_3_out/model906_30

Inference:

python detect.py --source xxx.mp4 --weights  xxx.pt --view-img

Training:

# training from scratch
python train.py --data data/match.yaml

# training from pretrain weight
python train.py --weights xxx.pt --data data/match.yaml

# resume training
python train.py --data data/match.yaml --resume

Evaluation:

python val.py --weights xxx.pt --data data/match.yaml

Deployment:

# Server
python deploy/app.py --weights xxx.pt

# Client
python deploy/test_app.py

Dataset Preparation:

# TrackNetV2 dataset
#	/home/chg/Badminton/TrackNetV2
#	- Amateur  
#	- Professional  
#	- Test

python tools/handle_tracknet_dataset.py /home/chg/Badminton/TrackNetV2/Amateur
python tools/handle_tracknet_dataset.py /home/chg/Badminton/TrackNetV2/Professional
python tools/handle_tracknet_dataset.py /home/chg/Badminton/TrackNetV2/Test

python tools/Frame_Generator_rally.py /home/chg/Badminton/TrackNetV2/Amateur
python tools/Frame_Generator_rally.py /home/chg/Badminton/TrackNetV2/Professional
python tools/Frame_Generator_rally.py /home/chg/Badminton/TrackNetV2/Test


# TrackNetV2 dataset config : data/match.yaml
path: /home/chg/Documents/Badminton/TrackNetV2
train:
    - Amateur
    - Professional 
val:
    - Test
    
# also you can use follow config for testing
train:
    - Test/match1/images/1_05_02
val:
    - Test/match2/images/1_03_03

# or
train:
    - Test/match1
val:
    - Test/match2

Reference：

https://github.com/mareksubocz/TrackNet

https://nol.cs.nctu.edu.tw:234/open-source/TrackNetv2

https://github.com/ultralytics/yolov5