CODEWITHSUNDEEP
pythonmachine-learningkerasautoencoder
veritas9872
veritas9872

Reputation: 99

Keras ImageDataGenerator and flow_from_directory class_mode='input'

I am trying to use Keras' ImageDataGenerator for a UNet autoencoder. I would like to import and convert RGB images from the Tiny ImageNet dataset into grayscale images which have been rescaled to have values between 0~1. 

train_dir = r'D:\tiny-imagenet-200-testing\tiny-imagenet-200\train'
train_datagen = ImageDataGenerator(rescale=1 / 255)
train_generator = train_datagen.flow_from_directory(train_dir, target_size=(64, 64), 
                   color_mode='grayscale', class_mode='input', batch_size=128, 
                   horizontal_flip=True)

I am trying to use the flow_from_directory with class_mode='input' in Keras, which says that:

"input" will be images identical to input images (mainly used to work with autoencoders). (See https://keras.io/preprocessing/image/)

However, I don't know whether the "input images" being returned are the rescaled and flipped images or the original data, unmodified by conditions specified in the ImageDataGenerator. Does anyone know the order in which the ImageDataGenerator and flow_from_directory interact with one another? Especially when class_mode='input'?

Upvotes: 2

Views: 4422

Answers (1)

today
today

Reputation: 33420

The generated images and their corresponding labels are the same in case of using class_mode='input'. You can confirm this by:

import numpy as np

for tr_im, tr_lb in train_generator:
    if np.all(tr_im == tr_lb):
        print('They are the same!`)
    break

The output of the above code would be They are the same!.

Upvotes: 3

Related Questions