rpi-rgb-led-matrix/python

Python bindings for RGB Matrix library

Building

In the root directory for the matrix library simply type:

Python 2

sudo apt-get update && sudo apt-get install python2.7-dev python-pillow -y
make build-python
sudo make install-python

Python 3

You can also build for Python 3:

sudo apt-get update && sudo apt-get install python3-dev python3-pillow -y
make build-python PYTHON=$(which python3)
sudo make install-python PYTHON=$(which python3)

PyPy

The cython binding to PyPy seems to be somewhat working but extremely slow (20x slower even than the regular Python binding, 160x slower than C++), so this is not recommended. From glancing at the docs, the library might need to be bound via CFFI first. If anyone has experience doing that, please consider preparing a pull request (ideally in a way that it can be used as well for Python2/3 and does not change the interface of the current Python binding).

Performance

The simplicity of scripting comes at a price: Python is slower than C++ of course. If you have to do a lot of pixel updates in your demo, this can be too slow depending on what you do. Here are some rough numbers for calling SetPixel() in a tight loop:

• On a Pi-2 and Pi-3, a Python script will be about 1/8 of the speed compared to the corresponding C++ program (pushing ~0.43 Megapixels/s Python vs. ~3.5 Megapixels/s C++ on a Pi-3 for instance)
• On a Pi-1, the difference is even worse: 1/24 of the speed to the corresponding C++ program. Given that the Pi-1 is already about 1/10 the speed of a Pi-3, this almost makes Python unusable on a Pi-1 (~0.015 Megapixels/s Python vs. ~0.36 Megapixels/s C++)
• Also interesting: Python3 is a little bit slower than Python2.7. So if you can, stick with Python2.7 for now.
• The good news is, that this is due to overhead per function call. If you can do more per function call, then this is less problematic. For instance if you have an image to be displayed with SetImage(), that will be faster per pixel. (Currently, SetImage() is still much slower than it could be, because it does not send the raw buffer into the C-side but does individual SetPixel() calls in the interface layer. If you are interested in improving the performance of Python, this would be a good start and welcomed pull request).

The ~0.015 Megapixels/s on a Pi-1 means that you can update a 32x32 matrix at most with ~15Hz. If you have chained 5, then you barely reach 3Hz. In a Pi-3, you get about 400Hz update rate (85Hz for 5-chain) with a Python program (while with C++, you can do the same thing with a comfortable 3500Hz (700Hz for 5)). Keep in mind that this is if all you do is just calling SetPixel(), it does not include any time of what you actually want to do in your demo - so anything in addition to that will drop your update rate.

Using the library

Be aware of the fact, that using the RGBMatrix requires root privileges. Therefore you will need to run all you python scripts as using sudo.

You find examples in the samples/ subdirectory. The examples all use the samplebase.py that provides some utility to all example programs, such as command-line parsing: all sample-programs accept --rows, --chain and --parallel as options to adapt to your configuration

cd samples
sudo ./runtext.py --chain=4

Here a complete example how to write an image viewer:

#!/usr/bin/env python
import time
import sys

from rgbmatrix import RGBMatrix
from PIL import Image

if len(sys.argv) < 2:
    sys.exit("Require an image argument")
else:
    image_file = sys.argv[1]

image = Image.open(image_file)

rows = 32
chain = 1
parallel = 1
matrix = RGBMatrix(rows, chain, parallel)

# Make image fit our screen.
image.thumbnail((matrix.width, matrix.height), Image.ANTIALIAS)

matrix.SetImage(image.convert('RGB'))

try:
    print("Press CTRL-C to stop.")
    while True:
        time.sleep(100)
except KeyboardInterrupt:
    sys.exit(0)