PyMosaic

Recreate a target images by making a mosaic out of multiple input images.

Examples

Note: Images have been scale down to take less space

Quickstart

To get started quickly you can use the provided example input and target images. The example is using the --grayscale option as it yields better results. Also, the --multithreading option is really important to get sufficient speed. The algorithm is quite long, and without multithreading you'll wait for a long time, especially if the image is big.

Simply run

make lena_grayscale_multithreading

or use the equivalent script call

python mosaic.py --target_im=images/target/lena.jpg			        \
	--inputs=images/input/ --resize_factor=1 --pixel_density=0.25	\
	--grayscale --output_size_factor=5 --contrast 					\
	--multithreading --num_workers=12 --save

To get a detailed list of the arguments, head to How to use section.

• --resize_factor=1 ensures the initial image size is kept the same when going through it
• --pixel_density=0.25 tells how big the pixel scanning must be
• --grayscale to use grayscale mode (yields better results)
• --output_size_factor=5 the output image will be 5 times the initial image size
• --contrast applies CLAHE to the image to enhance the constrat before using it

Resulting image

Note: Image has been scale down to take less space

How to use

You need to run python mosaic.py with the arguments described beneath. Take a look at the makefile to see well working examples.

Makefile

If you want to use the Makefile provided, please change the first line

python-bin = ~/.pyenv/versions/3.6.7/envs/pymosaic/bin/python

to your python bin path.

Required arguments

• --target_im : Path to target image
• --inputs : Path to input images

Optional arguments

• --output_size_factor : How much times the output should be bigger than the target
• --resize_factor : Factor to resize target image
• --grayscale : Convert to grayscale
• --contrast : Apply auto contrast to target image
• --color_filter : Apply color filters to get closer to the desired color
• --pixel_density : Path to target image
• --multithreading : Use multiple thread to create the mosaic
• --num_workers : Number of workers to use in multithreading
• --save : Save the output mosaic in results/
• --show : Show output mosaic

Recommended arguments

From what I've experienced, using --contrast in combination with --grayscale yields the best results. To keep the process short enough, use --multithreading with the highest --num_workers for your machine.

Don't forget --save to actually save the created mosaic!

How does it work?

I'm in the process of writing a post on my blog to give a detailed tutorial on that.

Requirements

Simply run

pip install -r requirements.txt

TODO

• Add argument parsing
• B&W support
• Be able to select output resolution
• Add option to save mosaic
• Add/test contrast increasing before mosaic
• Use pixel density argument
• Add color filtering option (for smoother results)
• Add option to choose to display the output or not
• [] Add script to download input images (from ImageNet)
• [] Handle case where kernel size = 1
• [] Handle case where kernel size is > 1 and < to kmean clusters (3)
• [] Improve color filtering
• [] Add web interface

Authors

• Victor MEUNIER - ZeroFrame - MrEliptik - [email protected]

Want to support me? Buy me a coffee!