YOLOv8 Custom Training: From Dataset to Production API

Dataset Preparation

1. Annotate with Roboflow or LabelImg
2. Export in YOLOv8 format
3. Data split: 80/10/10 train/val/test

Training

from ultralytics import YOLO

model = YOLO('yolov8n.pt')  # nano for fast prototyping
results = model.train(
    data='dataset.yaml',
    epochs=100,
    imgsz=640,
    batch=16,
    augment=True,
)

ONNX Export for Production

model.export(format='onnx', dynamic=True, simplify=True)

FastAPI Endpoint

@app.post('/detect')
async def detect(file: UploadFile):
    img = Image.open(file.file)
    results = model(img)
    return {'detections': results[0].boxes.json()}