Photos and Video Files

Detectors work on any image, not only a webcam.

Photos

Read the image with OpenCV, and set static_image=True so that each image is treated on its own:

import cv2 as cv
from detectors_world import create_detector

frame = cv.imread("photo.jpg")

with create_detector("hand", static_image=True) as hand:
    hands = hand.detect(frame)

cv.imwrite("result.jpg", frame)
print(f"Found {len(hands)} hands")

cv.imread() returns None when it can't read a file. Passing that to a detector raises a ValueError.

For a complete program with command line options, see Photo Detection.

A folder of photos

Reuse one detector, because creating a detector loads its model each time:

from pathlib import Path

with create_detector("face", static_image=True) as face:
    for path in Path("photos").glob("*.jpg"):
        frame = cv.imread(str(path))
        print(path.name, len(face.detect(frame, draw=False)))

Video files

Open the file with cv.VideoCapture, and leave static_image at its default so that the detector follows things between frames:

cap = cv.VideoCapture("clip.mp4")

with create_detector("pose") as pose:
    while True:
        ok, frame = cap.read()
        if not ok:
            break
        bodies = pose.detect(frame)

cap.release()

The Video File Processing example also saves the result to a new file.

When to use static_image

Your frames	Use
Webcam, or a video file	static_image=False (default)
Unrelated photos	static_image=True

In video mode, the model follows what it found in the previous frame, which is smoother and faster. On unrelated photos that gets in the way, so static_image=True searches each image from scratch.

Detecting several times on one frame

Each detect() call runs the model again. Call it once per frame and reuse the result, rather than calling it twice:

hands = hand.detect(frame)
first = hand.locate(frame, hand=0)
second = hand.locate(frame, hand=1)