new.py

# Install required package
# pip install ultralytics

from ultralytics import YOLO

# Dataset structure expected:
# ├── dataset/
# │   ├── train/
# │   │   ├── images/
# │   │   └── labels/
# │   ├── valid/
# │   │   ├── images/
# │   │   └── labels/
# │   └── test/
# │       ├── images/
# │       └── labels/

# data.yaml example:
# path: /path/to/dataset
# train: train/images
# val: valid/images
# test: test/images
# names:
#   0: class1
#   1: class2
#   ...

def train_yolov8():
    # Load the YOLOv8 Large model
    model = YOLO('yolov8l.pt')  # pretrained model

    # Train the model
    results = model.train(
        data='data.yaml',
        epochs=100,
        batch=16,
        imgsz=640,
        device='0',  # 'cpu' or '0' for GPU
        name='yolov8l_custom',
        optimizer='Adam',
        lr0=0.001,
        warmup_epochs=3,
        augment=True,
        patience=50,
        pretrained=True
    )

    # Validate the model
    metrics = model.val()  # Validate on validation set
    print(f"Validation [email protected]: {metrics.box.map}")

    # Test the model (optional)
    test_model = YOLO('runs/detect/yolov8l_custom/weights/best.pt')
    test_metrics = test_model.val(data='data.yaml', split='test')
    print(f"Test [email protected]: {test_metrics.box.map}")

    # Export to ONNX format (optional)
    model.export(format='onnx')

if __name__ == '__main__':
    train_yolov8()