Super Pyxelate converts images to 8-bit pixel art. It is an improved, faster implementation of the original Pyxelate algorithm with palette transfer support and enhanced dithering.
Super Pyxelate is currently in beta.
Usage
from skimage import io
from pyxelate import Pyx, Pal
# load image with 'skimage.io.imread()'
image = io.imread("examples/blazkowicz.jpg")
downsample_by = 14 # new image will be 1/14th of the original in size
palette = 7 # find 7 colors
# 1) Instantiate Pyx transformer
pyx = Pyx(factor=downsample_by, palette=palette)
# 2) fit an image, allow Pyxelate to learn the color palette
pyx.fit(image)
# 3) transform image to pixel art using the learned color palette
new_image = pyx.transform(image)
# save new image with 'skimage.io.imsave()'
io.imsave("pixel.png", new_image)
Pyxelate extends scikit-learn transformers, allowing the same learned palette to be reused on other, aesthetically similar images (so it's somewhat like an 8-bit style transfer):
car = io.imread("examples/f1.jpg")
robocop = io.imread("examples/robocop.jpg")
# fit a model on each
pyx_car = Pyx(factor=5, palette=8, dither="none").fit(car)
pyx_robocop = Pyx(factor=6, palette=7, dither="naive").fit(robocop)
"""
pyx_car.transform(car)
pyx_car.transform(robocop)
pyx_robocop.transform(car)
pyx_robocop.transform(robocop)
"""
For a single image, it is possible to call both fit() and transform() at the same time:
# fit() and transform() on image with alpha channel
trex = io.imread("examples/trex.png")
trex_p = Pyx(factor=9, palette=4, dither="naive", alpha=.6).fit_transform(trex)
Hyperparameters for Pyx()
| Parameter | Description |
|---|---|
| height | The height of the transformed image. If only height is set, the width of the transofmed image will be calculated to maintain the aspect ratio of the original. |
| width | The width of the transformed image. If only width is set, the height of the transofmed image will be calculated to maintain the aspect ratio of the original. |
| factor | The size of the transformed image will be 1. / factor of the original. Can be used instead of setting width or height. |
| upscale | Resizes the pixels of the transformed image by upscale. Can be a positive int or a tuple of ints for (h, w). Default is 1. |
| palette | The number of colors in the transformed image. - If it's an int that is larger than 2, Pyxelate will search for this many colors automatically. Default is 8. - If it's a Pal palette enum object, Pyxelate will use palette transfer to match these colors. |
| dither | The type of dithering to use on the transformed image (see more exampels below): - "none" no dithering is applied (default, takes no additional time)- "naive" Pyxelate's naive dithering based on probability mass function (use for images with alpha channel) - "bayer" Bayer-like ordered dithering using a 4x4 Bayer Matrix (fastest dithering method, use for large images)- "floyd" Floyd-Steinberg inspired error diffusion dithering (slowest)- "atkinson" Atkinson inspired error diffusion dithering (slowest) |
| alpha | For images with transparency, the transformed image's pixel will be either visible/invisible above/below this threshold. Default is 0.6. |
| sobel | The size of the sobel operator (N*N area to calculate the gradients for downsampling), must be an int larger than 1. Default is 3, try 2 for a much faster but less accurate output. |
| depth | How many times should the Pyxelate algorithm be applied to downsample the image. More iteratrions will result in blockier aesthatics. Must be a positive int, although it is really time consuming and should never be more than 3. Raise it only for really small images. Default is 1. |
| boost | Adjust contrast and apply preprocessing on the image before transformation for better results. In case you see unwanted dark pixels in your image set this to False. Default is True. |
Showcase of available dithering methods: 
See more examples in the example Jupyter Notebook.
Assigning existing palette
Common retro palettes are available in Pal:
from pyxelate import Pyx, Pal
vangogh = io.imread("examples/vangogh.jpg")
vangogh_apple = Pyx(factor=12, palette=Pal.APPLE_II_HI, dither="atkinson").fit_transform(vangogh)
vangogh_mspaint = Pyx(factor=6, palette=Pal.MICROSOFT_WINDOWS_PAINT, dither="none").fit_transform(vangogh)
Assign your own palette:
my_pal = Pal.from_hex(["#FFFFFF", "#000000"])
# same but defined with RGB values
my_pal = Pal.from_rgb([[255, 255, 255], [0, 0, 0]])
Fitting existing palettes on different images will also have different results for transform().
Installation
pip install git+https://github.com/sedthh/pyxelate.git --upgrade
Pyxelate relies on the following libraries to run (included in requirements.txt):
FAQ
The source code is available under the MIT license but I would appreciate the credit if your work uses Pyxelate (for instance you may add me in the Special Thanks section in the credits of your videogame)!
How does it work?
Pyxelate downsamples images by (iteratively) dividing it to 3x3 tiles and calculating the orientation of edges inside them. Each tile is downsampled to a single pixel value based on the angle the magnitude of these gradients, resulting in the approximation of a pixel art. This method was inspired by the Histogram of Oriented Gradients computer vision technique.
Then an unsupervised machine learning method, a Bayesian Gaussian Mixture model is fitted (instead of conventional K-means) to find a reduced palette. The tied gaussians give a better estimate (than Euclidean distance) and allow smaller centroids to appear and then lose importance to larger ones further away. The probability mass function returned by the uncalibrated model is then used as a basis for different dithering techniques.
Preprocessing and color space conversion tricks are also applied for better results.
PROTIPs
- There is no one setting fits all, try experimenting with different parameters for better results! A setting that generates visually pleasing result on one image might not work well for another.
- The bigger the resulting image, the longer the process will take. Note that most parts of the algorithm are O(H*W) so an image that is twice the size will take 4 times longer to compute.
- Assigning existing palettes will take longer for larger palettes, because LAB color distance has to be calculated between each color separately.
- Dithering takes time (especially atkinson) as they are mostly implemented in plain python with loops.
TODOs
- Add CLI tool for Pyxelate so images can be batch converted from command line.
- Re-implement Pyxelate for animations / sequence of frames in video.
- Include PIPENV python environment files instead of just setup.py.
- Implement Yliluoma's ordered dithering algorithm and experiment with improving visuals through gamma correction.
- Write a whitepaper on the Pyxelate algorithm.
![[Suggestion] Alternatives to hog for faster runtime](https://avatars.githubusercontent.com/u/46298009?v=4)
Factor 6
Factor 3
Factor 6
1 Oct 08, 2021
30 Dec 18, 2022
11 Dec 09, 2022
5 Oct 12, 2021
7 Dec 15, 2022
20 Nov 05, 2022
24 Apr 14, 2022
1 Feb 02, 2022
0 Aug 16, 2022
2k Dec 19, 2022
467 Dec 30, 2022
12 Jul 30, 2022
6 Nov 14, 2021
8 Dec 30, 2022
1 Oct 27, 2021
672 Jan 05, 2023
28 Mar 01, 2022
2 Jan 09, 2022
1.3k Jan 06, 2023
1.2k Dec 30, 2022