Example¶

There are three main classes:

• Master: which represents the master image.
• Pool: which is the pool of tile images to use to reconstruct the Master image.
• Mosaic: which is to construct the photo mosaic for a Master image and Pool tiles.

We'll be using the UTKFace dataset which can be downloaded here. Extract it to the faces/ folder.

It contains 9780 200x200 images of faces.

Let's import the 3 main classes.

In [1]:

from phomo import Master, Pool, Mosaic

from phomo import Master, Pool, Mosaic

/home/lcoyle/.cache/pypoetry/virtualenvs/phomo-pX3Qwu7w-py3.12/lib/python3.12/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html
  from .autonotebook import tqdm as notebook_tqdm

Pool¶

We initialize the pool from the faces/ folder. We'll be using 20x20 square tiles.

In [2]:

pool = Pool.from_dir("faces/", crop_ratio=1, tile_size=(20, 20))
pool

pool = Pool.from_dir("faces/", crop_ratio=1, tile_size=(20, 20)) pool

Loading tiles: 100%|█████████████████████████████████████████████████████████████████| 9780/9780 [00:04<00:00, 2102.21it/s]

Out[2]:

phomo.pool.Pool at 0x7fbf6e0436e0:
    len: 9780

Master¶

Let's take a random image from the faces/ folder and use that as the master image. We also resize the master image from 200x200 to 400x400.

In [3]:

from pathlib import Path

master_file = list(Path("faces").glob('*'))[1534]

from pathlib import Path master_file = list(Path("faces").glob('*'))[1534]

In [4]:

master = Master.from_file(master_file, img_size=(400, 400))
master

master = Master.from_file(master_file, img_size=(400, 400)) master

Out[4]:

phomo.master.Master at 0x7fbf6c91a6f0:
    shape: (400, 400, 3)

In [5]:

master.img

master.img

Out[5]:

Mosaic¶

We can create the Mosaic from the master image and the tile pool.

In [6]:

mosaic = Mosaic(master, pool)
mosaic

mosaic = Mosaic(master, pool) mosaic

Out[6]:

phomo.mosaic.Mosaic at 0x7fbf6e075af0:
    n_appearances: 1
    mosaic size: (400, 400)
    tile shape: (20, 20)
    leftover tiles: 9380
    phomo.grid.Grid at 0x7fbf55529700:
        origin: (0, 0)
        len slices: 400
        thresholds: []
    phomo.master.Master at 0x7fbf6c91a6f0:
        shape: (400, 400, 3)
    phomo.pool.Pool at 0x7fbf6e0436e0:
        len: 9780

Let's show the tile grid to see how things line up.

In [7]:

mosaic.grid.plot()

mosaic.grid.plot()

Out[7]:

If we want to get fancy we can subdivide tiles with more contrast. We'll get more detail in thoss tiles but have inconsistent tile size.

In [8]:

mosaic.grid.subdivide(0.1)
mosaic.grid.plot()

mosaic.grid.subdivide(0.1) mosaic.grid.plot()

Out[8]:

Next we compute the distance matrix, which will be used to decide which tile goes where. This is the process which takes the longest.

In [9]:

d_matrix = mosaic.d_matrix()

d_matrix = mosaic.d_matrix()

Building distance matrix: 100%|██████████████████████████████████████████████████████████| 493/493 [00:25<00:00, 18.96it/s]

If you have a GPU and installed phomo with the cuda extra (pip install 'phomo[cuda]'), you can use GPU acceleration to compute the distance metric.

In [10]:

# d_matrix = mosaic.d_matrix_cuda()

# d_matrix = mosaic.d_matrix_cuda()

Now we give the distance matrix to the mosaic to build the mosaic.

In [11]:

mosiac_img = mosaic.build(d_matrix)
mosiac_img

mosiac_img = mosaic.build(d_matrix) mosiac_img

Building mosaic: 100%|████████████████████████████████████████████████████████████████| 493/493 [00:00<00:00, 95864.25it/s]

Out[11]:

If we want, we can use PIL to blend the master with the mosaic to smooth out some of the jaggedness.

In [12]:

from PIL import Image

Image.blend(mosiac_img, master.img, 0.3)

from PIL import Image Image.blend(mosiac_img, master.img, 0.3)

Out[12]: