Skip to content

Instantly share code, notes, and snippets.

@chsasank
Forked from fmder/elastic_transform.py
Last active October 14, 2023 01:55
Show Gist options
  • Save chsasank/4d8f68caf01f041a6453e67fb30f8f5a to your computer and use it in GitHub Desktop.
Save chsasank/4d8f68caf01f041a6453e67fb30f8f5a to your computer and use it in GitHub Desktop.
Elastic transformation of an image in Python
import numpy as np
from scipy.ndimage.interpolation import map_coordinates
from scipy.ndimage.filters import gaussian_filter
def elastic_transform(image, alpha, sigma, random_state=None):
"""Elastic deformation of images as described in [Simard2003]_.
.. [Simard2003] Simard, Steinkraus and Platt, "Best Practices for
Convolutional Neural Networks applied to Visual Document Analysis", in
Proc. of the International Conference on Document Analysis and
Recognition, 2003.
"""
assert len(image.shape)==2
if random_state is None:
random_state = np.random.RandomState(None)
shape = image.shape
dx = gaussian_filter((random_state.rand(*shape) * 2 - 1), sigma, mode="constant", cval=0) * alpha
dy = gaussian_filter((random_state.rand(*shape) * 2 - 1), sigma, mode="constant", cval=0) * alpha
x, y = np.meshgrid(np.arange(shape[0]), np.arange(shape[1]), indexing='ij')
indices = np.reshape(x+dx, (-1, 1)), np.reshape(y+dy, (-1, 1))
return map_coordinates(image, indices, order=1).reshape(shape)
@koegl
Copy link

koegl commented Jun 17, 2022

Sorry about my poor knowledge, but you are using gaussian filter for what exactly?

It is used to create the random deformations dx and dy. If you didn't have dx and dy, then the deformed mesh grid would be the same as the un-deformed mesh grid - so no deformation would happen

@koegl
Copy link

koegl commented Jun 17, 2022

for those who have trouble with using this code
first, the image has to be square like: (x,x,3)
second, for showing the transformed image ex: transformed_image=elastic_transform(img,40,2) then using plt.imshow(transformed_image)

  1. The image does not have to be square
  2. The image has to be grayscale, i.e. the shape has to be (a,b) - this line of code makes sure of that assert len(image.shape)==2

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment