TechOverflow Logo
TechOverflow

Is pypng 16-bit PNG encoding faster using pypy on the Raspberry Pi?

In our previous post How to save Raspberry Pi raw 10-bit image as 16-bit PNG using pypng we investigated how to use the pypng library to save 10-bit raw Raspberry Pi Camera images to 16-bit PNG files.

However, saving a single image took ~26 seconds using CPython 3.7.3. Since pypy can provide speedups to many Python workloads, we tried using pypy3 7.0.0 (see How to install pypy3 on the Raspberry Pi) to speed up the PNG encoding.

Results

pypng PNG export seems to be one of the workloads that are much slower using pypy3.

Encoding is more that 10x slower when using pypy3!

Hence I don’t recommend using pypy3 to speed up pypng encoding workloads, at least not on the Raspberry Pi!

Full example

This example is derived from our full example previously posted on How to save Raspberry Pi raw 10-bit image as 16-bit PNG using pypng:

#!/usr/bin/env python3
import time
import picamera
import picamera.array
import numpy as np
import png

# Capture image
print("Capturing image...")
with picamera.PiCamera() as camera:
    with picamera.array.PiBayerArray(camera) as stream:
        camera.capture(stream, 'jpeg', bayer=True)
        # Demosaic data and write to rawimg
        # (stream.array contains the non-demosaiced data)
        rawimg = stream.demosaic()

# Write to PNG
print("Writing 16-bit PNG...")
t0 = time.time()
with open('16bit.png', 'wb') as outfile:
    writer = png.Writer(width=rawimg.shape[1], height=rawimg.shape[0], bitdepth=16, greyscale=False)
    # rawimg is a (w, h, 3) RGB uint16 array
    # but PyPNG needs a (w, h*3) array
    png_data = np.reshape(rawimg, (-1, rawimg.shape[1]*3))
    # Scale 10 bit data to 16 bit values (else it will appear black)
    # NOTE: Depending on your photo and the settings,
    #  it might still appear quite dark!
    png_data *= int(2**6)
    writer.write(outfile, png_data)
t1 = time.time()

print(f"Encoding took {(t1 - t0):.2f} seconds")