YOLOv5 implementation using TensorFlow 2

Train

• Change data_dir, image_dir, label_dir and class_dict in config.py
• Choose version in config.py
• Optional, python main.py --anchor to generate anchors for your dataset and change anchors in config.py
• Optional, python main.py --record to generate tf-record for your dataset
• Run python main.py --train for training

Test

• Run python main.py --test

Dataset structure

├── Dataset folder 
    ├── images
        ├── 1111.jpg
        ├── 2222.jpg
    ├── labels
        ├── 1111.xml
        ├── 2222.xml
    ├── train.txt
    ├── test.txt

Note

• xml file should be in PascalVOC format
• train.txt test.txt contains image names without extension

Recommendation (for docker users)

• docker pull nvcr.io/nvidia/tensorflow:20.12-tf2-py3
• nvidia-docker run --gpus all -v /your/project/folder:/Projects -it nvcr.io/nvidia/tensorflow:20.12-tf2-py3
• cd ../Projects
• apt-get update
• apt-get install ffmpeg libsm6 libxext6 -y
• pip install opencv-python

Reference

• https://github.com/ultralytics/yolov5
• https://github.com/wizyoung/YOLOv3_TensorFlow