opencv-yolov3-multiprocess-test.py

#!/usr/bin/env python3

import cv2 as cv
import multiprocessing


def do_detection(backend, process_type):
    print(f"Starting detection on {backend} from {process_type}")

    net = cv.dnn.readNet(model="yolov3.weights", config="yolov3.cfg")

    if backend == "CUDA":
        net.setPreferableBackend(cv.dnn.DNN_BACKEND_CUDA)
        net.setPreferableTarget(cv.dnn.DNN_TARGET_CUDA)
    else:
        net.setPreferableBackend(cv.dnn.DNN_BACKEND_OPENCV)
        net.setPreferableTarget(cv.dnn.DNN_TARGET_CPU)

    image = cv.imread("lenna.png")

    blob = cv.dnn.blobFromImage(image, 1 / 255, (416, 416), (0, 0, 0))

    net.setInput(blob)
    detections_blob = net.forward()

    print(f"Results of detection on {backend} from {process_type}\n{detections_blob}\n\n")


def start_detection_multiprocess(backend):

    process = multiprocessing.Process(target=do_detection, args=(backend, "multiprocessing"))

    process.start()
    process.join()



if __name__ == '__main__':
    multiprocessing.set_start_method('spawn')
    
    do_detection("CPU", "mainprocess")
    do_detection("CUDA", "mainprocess")

    start_detection_multiprocess("CPU")
    start_detection_multiprocess("CUDA")

nice. I got the cuda backend working without complaints in python, except it is slow (making me suspect it is not actually using it).