Yolov7App

Overview

Yolov7App is a toolkit for object detection using the YOLOv7 model. It is derived from an open-source repository and is maintained to provide a complete workflow for data processing, model training, ONNX model conversion, and C++ deployment.

Installation

1. Clone the repository

git clone https://github.com/GilbertPan97/Yolov7App
cd Yolov7App

2. Install dependencies

pip install -r requirements.txt

Usage Manual

1. Data Conversion (Labelme to YOLO TXT Format)

Ensure your Labelme dataset is organized and use the provided script for conversion.

python tools/convert_labelme_to_txt.py --input_folder <input-folder> \
    --output_folder <output-folder> --label_map <categories-json> \
    --image_extension <image-format> --train_ratio <train-ratio> --val_ratio <val-ratio> --test_ratio <test-ratio>

• <input-folder>: Directory containing images and LabelMe JSON annotation files.
• <output-folder>: Directory to save the converted YOLO TXT files.
• <categories-json>: Path to the category mapping JSON file.
• <image_extension>: Image file format (e.g., png, jpg).
• <train-ratio>: Ratio of the dataset for training.
• <val-ratio>: Ratio of the dataset for validation.
• <test-ratio>: Ratio of the dataset for testing.

The script will generate .txt label files where each line corresponds to an object's class and bounding box in YOLO format.

2. Model Training

After preparing the data, start training with the following command:

python train.py --workers 8 --device 0 --batch-size 10 \
    --data <data-config.yaml> --img 640 640 \
    --cfg <model-config.yaml> --weights <initial-weights.pt> \
    --name <experiment-name> --hyp <hyperparameter-config.yaml>

• <data-config.yaml>: Path to the dataset configuration file.
• <model-config.yaml>: Path to the model configuration file.
• <initial-weights.pt>: Path to the initial weights (or use 'yolov7.pt' for pretrained weights).
• <experiment-name>: Name for the training experiment.
• <hyperparameter-config.yaml>: Path to the hyperparameter configuration file.

Example:

python train.py --workers 8 --device 0 --batch-size 10 \
    --data /media/gilbertpan/Elements/AlgaeDataset/WorkSpace/dataset_config.yaml --img 640 640 \
    --cfg cfg/training/yolov7.yaml --weights 'runs/train/box/weights/best.pt' \
    --name yolov7 --hyp data/hyp.scratch.p5.yaml

3. Model Conversion (to ONNX Format)

Once training is complete, convert the model to ONNX format for deployment.

python export.py --weights <trained-weights.pt>

Example:

python export.py --weights ./runs/train/yolov72/weights/best.pt

The exported ONNX model will be saved in the runs/train/exp/weights directory.

Notes

• Ensure that the directory structure is correctly set up for training.
• Adjust hyperparameters in the configuration files as needed for your specific dataset.

License

This project is licensed under the GNU General Public License.

Acknowledgments

Special thanks to the original authors of the YOLOv7 project for their outstanding work and contribution to the computer vision community.

Contact

For any questions or issues, please contact gilbertpan97@gmail.com.