[go: up one dir, main page]

Skip to content

Tools used for aspect-ratio-preserving image augmentation, namely translate, rotate, crop WITHOUT stretching or skewing the image, or padding pixels to fill up empty space.

License

Notifications You must be signed in to change notification settings

saravanabalagi/imaugtools

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

19 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

imaugtools

imaugtools contains tools used for image augmentation: translate, rotate, crop. This library is for you if you do NOT want to stretch or skew, or pad pixels that would make your images look strange when doing any of these operations.

Installation

To install imaugtools, simply run in your terminal

pip install imaugtools

Usage

Example image

example-image

Translation

from imaugtools import translate_image
my_image_translated = translate_image(my_image, 0.1, 0.2)
my_image_translated_uncropped = translate_image(my_image, 0.1, 0.2, crop=False)
imshow(my_image_translated, my_image_translated_uncropped, mode='BGR')

imshow function used here is to simply display the images, you can use it from imshowtools library.

With and without crop:

translated-image

Rotation

from imaugtools import rotate_image
my_image_rotated = rotate_image(my_image, 30) # angle in degrees
my_image_rotated_uncropped = rotate_image(my_image, 30, crop=False)
imshow(my_image_rotated, my_image_rotated_uncropped, mode='BGR')

With and without crop:

rotated-image

Cropping

from imaugtools import center_crop, crop_around_center
my_image_center_cropped = center_crop(my_image, (150, 200))
my_image_cropped_around_center = crop_around_center(my_image, (150, 200))
imshow(my_image_center_cropped, my_image_cropped_around_center, mode='BGR')

Aspect ratio preserving center crop and crop around center:

cropped-image

print(my_image_center_cropped.shape, my_image_cropped_around_center.shape)
# Output: (150, 200, 3) (150, 200, 3)

Advanced Usage

Strided Translation

Strided translation is one of the powerful image augmentation techniques used in training neural networks.

tx_max = 1 and ty_max = 1 is equivalent to a stride of 1 in both directions. After you specify tx_max, you can specify tx (translation in x-axis) from -tx_max to +tx_max. The same applies to ty and ty_max.

my_images_translated = []
for j in range(-1, 2):
    for i in range(-1, 2):
        my_images_translated.append(translate_image(my_image, i, j, tx_max=1, ty_max=1))
imshow(*my_images_translated, mode='BGR')

stride-1-translation

tx_max = 0.5 and ty_max = 0.5 is equivalent to a stride of 0.5 in both directions

my_images_translated = []
for j in range(-2, 3):
    for i in range(-2, 3):
        my_images_translated.append(translate_image(my_image, i/4, j/4, tx_max=0.5, ty_max=0.5))
imshow(*my_images_translated, mode='BGR')

stride-0.5-translation

Contributing

Pull requests are very welcome.

  1. Fork the repo
  2. Create new branch with feature name as branch name
  3. Check if things work with a jupyter notebook
  4. Raise a pull request

Licence

Please see attached Licence

About

Tools used for aspect-ratio-preserving image augmentation, namely translate, rotate, crop WITHOUT stretching or skewing the image, or padding pixels to fill up empty space.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published