Images

Run inference on a single image using a URL, file path, or NumPy array. This page shows how to use each input type with a loaded model.

Estimated read time: 2 minutes

The simplest way to run inference on an image is to call the model like a function. Model objects are callable—they implement __call__, which delegates to predict. In other words:

result = model(
    image_source
)  # exactly the same as: result = model.predict(image_source)

This page shows how to run inference on three common input types: a URL, a local file path, and a NumPy array.

Common setup (used in all cases)

from degirum_tools import ModelSpec, Display

# Configure & load once
model_spec = ModelSpec(
    model_name="yolov8n_coco--640x640_quant_hailort_multidevice_1",
    zoo_url="degirum/hailo",
    inference_host_address="@local",
    model_properties={"device_type": ["HAILORT/HAILO8", "HAILORT/HAILO8L"]},
)
model = model_spec.load_model()

Image URL

from degirum_tools import remote_assets  # small gallery of public sample URLs

image_url = remote_assets.three_persons  # or any reachable image URL (string)
result = model(image_url)

with Display("AI Camera — URL") as output_display:
    output_display.show_image(result.image_overlay)

URLs behind authentication or hosted on an intranet may fail if your runtime can't access them.

Local file path

from pathlib import Path

image_path = Path.home() / "Pictures" / "test.jpg"  # change as needed
result = model(str(image_path))  # file paths are accepted

with Display("AI Camera — File") as output_display:
    output_display.show_image(result.image_overlay)

Windows: Use raw strings for paths, e.g., r"C:\Users\you\Pictures\test.jpg".

NumPy array

# pip install opencv-python
import cv2

frame_bgr = cv2.imread("test.jpg")  # returns BGR ndarray
if frame_bgr is None:
    raise FileNotFoundError("Could not read test.jpg")

result = model(frame_bgr)

with Display("AI Camera — Array") as output_display:
    output_display.show_image(result.image_overlay)

That’s it—choose the type that matches your input, reuse the common setup, and call model(...) or model.predict(...) to get results.