predict.py

from ultralytics import YOLO
from PIL import Image
import cv2

model_path = "D:/projects/pyprojects/yolo-detection/runs/detect/train3/weights/best.onnx"

print("model path={}".format(model_path))

model = YOLO(model_path)
# accepts all formats - image/dir/Path/URL/video/PIL/ndarray. 0 for webcam
results = model.predict(source="0") # from video
#results = model.track(source="0", show=True)
#results = model.predict(source="folder", show=True) # Display preds. Accepts all YOLO predict arguments

# from PIL
#im1 = Image.open("bus.jpg")
#results = model.predict(source=im1, save=True)  # save plotted images

# from ndarray
#im2 = cv2.imread("bus.jpg")
#results = model.predict(source=im2, save=True, save_txt=True)  # save predictions as labels

# from list of PIL/ndarray
#results = model.predict(source=[im1, im2])