CVUtils - Computer Vision Utilities
The library of convenience utilities for computer vision experiments
It is the code which I wrote during my CV experiments to simplify some repeating procedures. The code is quite old, dirty and incomplete, but may be a starting point for something more.
Clone the repo and use as regular python package.
import CVUtils as cvu
images = cvu.imread('path/to/image/folder')
cvu.imshow(images)cvu.imshow(image)This function allows to plot the image using matplotlib, but is more convenient.
First of all, you can plot several images with the same call (unfortunately does not support different image shapes yet)
cvu.imshow(images)where images is array of shape (n, h, w, c)
... or path to the folder
figsize and images layout are computed automatically: images are trying to form the layout that is closest to the golden ratio for better view
However, you can customize the layout and the plot
cvu.imshow(images,
shape=(3, 2),
figsize=(2, 4),
titles=[str(i) for i in range(len(images))],
suptitle='Example')
You can read image, or whole image folder, or a part of it using max_img parameter.
images = cvu.imread('path/to/images', max_img=100)Common preprocessing functions are implemented like:
img = cvu.imread('some_example.png')
img_norm = cvu.normalize(img) # subtract mean and divide by std
img_float = cvu.to_1(img_norm) # normalize to the range [0, 1]
img_int = cvu.to_255(img_float) # shrink into [0, 255] dtype=uint8Advanced resize can be used like:
cvu.resize(img, (224, 224)) # simple resize
cvu.resize(img, (224, 224), save_aspect=True) # by long side
cvu.resize(img, (224, 244), save_aspect=True, pad=(0, 0, 0)) # pads the border blackYou can find more in the code - it is documented.