Python multiprocess with keras and opencv

Question

I am trying to run 2 processes in parallel using Python multiprocess but the second process always hangs up. I saw in other posts that importing the keras library inside the function solves the problem but it didn't work for me. Can't figure out what I am missing... Here is a short summary of my code:

import multiprocessing
import numpy as np


def detect_img():
    from keras.layers import Dense
    from keras.models import Sequential

    #Not the actual model
    model = Sequential()
    model.add(Dense(15, input_shape=(4,), activation='relu'))
    model.add(Dense(15, activation='relu'))
    model.add(Dense(15, activation='relu'))
    model.add(Dense(20, activation='relu'))
    model.add(Dense(4, activation='linear'))
    while True:
        x = np.random.rand(1, 4)
        y = model.predict(x)
        print(y)


def show_cam():
    import cv2
    cap = cv2.VideoCapture(0)
    while True:
        ret, frame = cap.read()
        cv2.imshow('frame', frame)
        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()

if __name__ == '__main__':

    p1 = multiprocessing.Process(target=show_cam())
    p2 = multiprocessing.Process(target=detect_img())

    p1.start()
    p2.start()

Answer 1

Ok. I faced quite similar behaviour in many experiments. After looking in the web I found a post explaining that opencv implements its own parallelization interfering with python GIL (Global Interpreter Lock) (I maybe can go into details if you like ... ). The solution that worked for me was to avoid using cv2!

Of course, this is not the best approach atm. but there is not enough time to look for a workaround ...