Documentation | Tutorials | Release Notes | 中文

tfyolo is a YOLO (You only look once) library implemented by TensorFlow2

Key Features

• minimal Yolov5 by pure tensorflow2
• yaml file to configure the model
• custom data training
• mosaic data augmentation
• label encoding by iou or wh ratio of anchor
• positive sample augment
• multi-gpu training
• detailed code comments
• full of drawbacks with huge space to improve

Tutorial

prepare the data

$ bash data/scripts/get_voc.sh
$ cd yolo
$ python dataset/prepare_data.py

Clone and install requirements

$ git clone git@github.com:LongxingTan/Yolov5.git
$ cd Yolov5/
$ pip install -r requirements.txt

Train

$ python train.py

Inference

$ python detect.py
$ python test.py

Train on custom data

If you want to train on custom dataset, PLEASE note the input data should like this:

image_dir/001.jpg x_min, y_min, x_max, y_max, class_id x_min2, y_min2, x_max2, y_max2, class_id2

And maybe new anchor need to be created, don't forget to change the nc(number classes) in yolo-yaml.

$ python dataset/create_anchor.py

Performance

Citation

If you find tf-yolo project useful in your research, please consider cite:

@misc{tfyolo2021,
    title={TFYOLO: yolo series benchmark in tensorflow},
    author={Longxing Tan},
    howpublished = {\url{https://github.com/longxingtan/tfyolo}},
    year={2021}
}