Keras ImageDataGenerator: random transform

user_6131248 0
Viewed 3888

I'm interested in augmenting my dataset with random image transformations. I'm using Keras ImageDataGenerator, and I'm getting the following error when trying to apply random_transform to a single image:

--> x = apply_transform(x, transform matrix, img_channel_axis, fill_mode, cval)
>>> RuntimeError: affine matrix has wrong number of rows.

I found the source code for the ImageDataGenerator here. However, I'm not sure how to debug the runtime error. Below is the code I have:

from keras.preprocessing.image import img_to_array, load_img
from keras.preprocessing.image import ImageDataGenerator
from keras.applications.inception_v3 import preprocess_input    

image_path = './figures/zebra.jpg'

#data augmentation
train_datagen = ImageDataGenerator(
    rotation_range=40,
    width_shift_range=0.2,
    height_shift_range=0.2,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True,
    fill_mode='nearest')

print "\nloading image..."
image = load_img(image_path, target_size=(299, 299))
image = img_to_array(image)
image = np.expand_dims(image, axis=0) # 1 x input_shape
image = preprocess_input(image)

train_datagen.fit(image)
image = train_datagen.random_transform(image)

The error occurs at the last line when calling random_transform.

keras deep-learning computer-vision
1 Answers

The problem is that random_transform expects a 3D-array.

See the docstring:

def random_transform(self, x, seed=None):
    """Randomly augment a single image tensor.
    # Arguments
        x: 3D tensor, single image.
        seed: random seed.
    # Returns
        A randomly transformed version of the input (same shape).
    """

So you'll need to call it before np.expand_dims.

user_15314630
Related