Hfrxdjt

I've written the following class that generates augmented images one by one. However, I'd like to be able to generate batches of 8 images each (or any number really). The way I'm hoping to do it is:

1. Create batches of resized images (store them in a tensor, perhaps?)


2. Iterate through batches (tensors?) and apply augmentation (the augmenter would return the augmented images)


3. Add the augmented images to a batch (new tensor) of their own and yield this new batch

Is this a valid approach and if so, how could I go about doing this? And if not, how else might it be done?

class ImagePreprocessor(object):

 def __init__(self, path_to_input_images):

 if not os.path.exists(path_to_input_images):
 raise ValueError('path(s) doesn't exist!')

 self.input_root, self.input_folders, self.input_files = next(os.walk(path_to_input_images))

 def load_image_from_file(self, filename):
 filepath = os.path.join(self.input_root, filename)
 img = skio.imread(filepath) # returns image as ndarray
 return img

 def create_augmenter(self):
 # (augmenter details removed for brevity; I know the augmenter works)
 return augmenter

 def augment(self, img):
 augmenter = self.create_augmenter()
 augmented_image = augmenter.augment_image(img)
 return augmented_image

 def generate_data(self):
 images = self.input_files

 i = 0
 shuffle(images)
 while True:
 # If all the images have been gone through, shuffle and restart
 if i >= len(images):
 i = 0
 shuffle(images)


 # Here is where I would like to create batches of images so that
 # I can yield entire batches rather than just one image at a time

 im = self.load_image_from_file(images[i])
 resized_im = cv2.resize(im,(2000,2000))
 aug_img = self.augment(resized_im)
 i += 1
 yield aug_img

```