Home

Awesome

patchify

patchfy can split images into small overlappable patches by given patch cell size, and merge patches into original image.

This library provides two functions: patchify, unpatchify.

Installation

pip install patchify

Usage

Split image to patches

patchify(image_to_patch, patch_shape, step=1)

2D image:

#This will split the image into small images of shape [3,3]
patches = patchify(image, (3, 3), step=1)

3D image:

#This will split the image into small images of shape [3,3,3]
patches = patchify(image, (3, 3, 3), step=1)

Merge patches into original image

unpatchify(patches_to_merge, merged_image_size)

reconstructed_image = unpatchify(patches, image.shape)

This will reconstruct the original image that was patchified in previous code.

Help! unpatchify yields distorted images

In order for unpatchify to work, patchies should be created with equal step size. e.g. if the original image has width 3 and the patch has width 2, you cannot really create equal step size patches with step size 2. (first patch [elem0, elem1] and second patch [elem2, elem3], in which elem3 is out of bound).

The required condition to successfully recover the image using unpatchify is to have (width - patch_width) mod step_size = 0 when calling patchify.

Full running examples

2D image patchify and merge

import numpy as np
from patchify import patchify, unpatchify

image = np.array([[1,2,3,4], [5,6,7,8], [9,10,11,12]])

patches = patchify(image, (2,2), step=1) # split image into 2*3 small 2*2 patches.

assert patches.shape == (2, 3, 2, 2)
reconstructed_image = unpatchify(patches, image.shape)

assert (reconstructed_image == image).all()

3D image patchify and merge

import numpy as np
from patchify import patchify, unpatchify

image = np.random.rand(512,512,3)

patches = patchify(image, (2,2,3), step=1) # patch shape [2,2,3]
print(patches.shape) # (511, 511, 1, 2, 2, 3). Total patches created: 511x511x1

assert patches.shape == (511, 511, 1, 2, 2, 3)
reconstructed_image = unpatchify(patches, image.shape)
print(reconstructed_image.shape) # (512, 512, 3)

assert (reconstructed_image == image).all()