How to pass an image path as parameter and only preprocess it?

[Kongsea]

It seems Augmentor needs a work directory as the parameter and it will preprocess all files in it.
However, we always need to preprocess a specific image and use the image path as the parameter.
So how to use Augmentor in this situation?
Thank you.

[mdbloice]

Hi there, the best thing to do would be to pass the single image as an array. So create a pipeline object without specifying an image directory:

import Augmentor
import numpy as np
from PIL import Image
# Create a Pipeline object without specifying a directory:
p = Augmentor.Pipeline()
# Add operations to the pipeline object as per usual:
p.rotate(probability=1, max_left_rotation=5, max_right_rotation=5)

and then pass an image as an array to the function sample_with_array():

img = Image.open("/tmp/test.JPEG")
img_array = np.asarray(img)
img_augmented = p.sample_with_array(img_array)

The image contained in img_augmented is an image in PIL format.

Admittedly this is not very intuitive, I'll try to fix that for a future version. Also the documentation for this functionality is sparse, I'll fix that too in an upcoming version.

M.

[Kongsea]

Thank you.
Look forward to the new version.

[mhnatiuk]

Hi,
I need to use Augmentor to process images file-by-file (i am generating artifical dataset and I have to do more operations than just augmenting). Im doing exactly what you posted above, however for
augmentation_pipeline.sample_with_array(np.asarray(img)) (a numpy array) I get error:

    return images[0]  # Here we return only the first image for the generators.
IndexError: list index out of range

[unnir]

@mhnatiuk

you can do it like this:

import Augmentor
import numpy as np
from PIL import Image
# Create a Pipeline object without specifying a directory:
p = Augmentor.Pipeline()
# Add operations to the pipeline object as per usual:
p.rotate(probability=1, max_left_rotation=5, max_right_rotation=5)

And then change your image:
p.sample_with_array((X_train[0].reshape(480,480)))

If it does not work, check this:
#90

[ddfabbro]

I used p._execute_with_array instead

img = Image.open("/tmp/test.JPEG")
img_array = np.asarray(img)
img_augmented = p._execute_with_array(img_array)