Cython + OpenCV and NumPy

Question

I have a program with mainly OpenCV and NumPy, with some SciPy as well. The system needs to be a real-time system with a frame rate close to 30 fps but right now only about 10 fps. Will using Cython help speed this up? I ask because OpenCV is already written in C++ and should already be quite optimized, and NumPy, as far as I understand, is also quite optimized. So will the use of Cython help improve the processing time of my program?

Answer 1

Hope this helps someone

Found this awesome post Use Cython to get more than 30X speedup on your Python code

Used the same factorial calculation inside video stream through the camera for two frames each

video_python.py

import numpy as np
import cv2
import time

def function(number):
    
    cap = cv2.VideoCapture(0)

    increment = 0

    while(True):
        # Capture frame-by-frame
        ret, frame = cap.read()

        # Our operations on the frame come here
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

        # Display the resulting frame
        cv2.imshow('frame',gray)

        start_time = time.time()
        y = 1
        for i in range(1, number+1):
            y *= i

        increment+=1
        if increment >2:
            # print(time.time()-start_time)
            print('Python increment ',increment)
            break
        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()
    return 0

video_cython.pyx

import numpy as np
import cv2
import time

cpdef int function(int number):

    cdef bint video_true = True
    
    cap = cv2.VideoCapture(0)

    cdef int y = 1
    cdef int i

    cdef int increment = 0
    cdef int increment_times = 0

    while(video_true):
        # Capture frame-by-frame
        ret, frame = cap.read()

        # Our operations on the frame come here
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

        # Display the resulting frame
        cv2.imshow('frame',gray)

        start_time = time.time()


        for i in range(1, number+1):
            y *= i
        increment_times+=1

        if increment_times > 2:
            # print(time.time()-start_time)
            print('Cython increment ',increment_times)
            break

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break

    cap.release()
    cv2.destroyAllWindows()
    return 0

setup.py

from distutils.core import setup
from Cython.Build import cythonize

setup(ext_modules = cythonize('video_cython.pyx',compiler_directives={'language_level' : "3"}))

then run

python setup.py build_ext --inplace

video_test.py

import video_python
import video_cython
import time
number = 100000

start = time.time()
video_python.function(number)
end =  time.time()

py_time = end - start
print("Python time = {}".format(py_time))

start = time.time()
video_cython.function(number)
end =  time.time()

cy_time = end - start
print("Cython time = {}".format(cy_time))

print("Speedup = {}".format(py_time / cy_time))

result:

Python increment 3

Python time = 6.602917671203613

Cython increment 3

Cython time = 0.4903101921081543

Speedup = 13.466817083311046

So doing any kind of python related stuff inside the loop can increase speed

Answer 2

Using Cython won't make a significant difference in this problem.

To get a profile/benchmark of your code the Pycharm IDE has a profiling tool, or you can use kernprof.

However, as a test, you can convert your code to the Cython code or C code with these instructions:

• Compile main Python program using Cython.
• Specifically, convert python code to C code.

---

[NOTE]:

This approach is for Python3, but it can also be applied to Python2.7 with a few changes. I've tested it before.

---

[UPDATE]:

You can also use PyInstaller and Nuitka to test another way to convert your code as compiled executable.