For object detection tasks in PyTorch using YOLOv8, the Ultralytics library provides a streamlined approach. Begin by ensuring the environment supports GPU acceleration if available.

import torch
from ultralytics import YOLO
import os

os.environ['KMP_DUPLICATE_LIB_OK'] = 'True'

def setup_model():
    device_type = 'cuda' if torch.cuda.is_available() else 'cpu'
    print(f"Using device: {device_type}")
    
    model = YOLO('yolov8n.pt').to(device_type)
    return model, device_type

def train_model(model, config_path, device_type):
    model.train(data=config_path, epochs=100, device=device_type, workers=0, batch=2)
    model.val(device=device_type)

if __name__ == '__main__':
    yolo_model, device = setup_model()
    train_model(yolo_model, 'dataset_config.yaml', device)

Model weights (yolov8n.pt) should be downloaded from the official Ultralytics documentation and placed in the working directory. Additionally, install the ultralytics package via pip.

Dataset configuration is managed through a YAML file that defines paths to training, valdiation, and test sets:

train: /path/to/training/data
val: /path/to/validation/data
test: /path/to/test/data

nc: 3
classes: [rice, animal, other]

The dataset must follow standard YOLO formatting where each image has a corresponding annotation file with matching filenames. Tools like makesense.ai can assist in creating labeled datsaets.

After training, best performing model is saved as best.pt within the generated runs/detect/train/weights/ directory. To perform inference:

import matplotlib.pyplot as plt
from ultralytics import YOLO

inference_model = YOLO('runs/detect/train/weights/best.pt')
sample_image = 'test_images/sample.jpg'

predictions = inference_model(sample_image)

if isinstance(predictions, list):
    for prediction in predictions:
        prediction.show()
else:
    predictions.show()

plt.imshow(predictions[0].orig_img)
plt.axis('off')
plt.show()

This workflow enables rapid deployment of custom object detection models using YOLOv8 architecture.