Tensorflow image_dataset_from_directory for input dataset and output dataset

Question

I am trying to learn image auto encoding but I can't use input and output images to train the model

ex: input images folder: ".../Pictures/Input"
output images folder: ".../Pictures/Output"

#get input images from data_dir_input
ds_input = tf.keras.preprocessing.image_dataset_from_directory(
    data_dir_input,
    seed=123,
    image_size=(img_height, img_width),
    label_mode=None,
    batch_size=batch_size)

#get output images from data_dir_output
ds_output = tf.keras.preprocessing.image_dataset_from_directory(
    data_dir_output,
    seed=123,
    image_size=(img_height, img_width),
    label_mode=None,
    batch_size=batch_size)

# --------- model init etc --------------
# ...


model.fit(x=ds_input, y=ds_output, batch_size=32, epochs=50)

But I get error saying this:

`y` argument is not supported when using dataset as input

How can i use my own input images and output images when training model?

Answer 1

You could use tf.data.Dataset for some more flexibility. From what I read, image_dataset_from_directory doesn't support any custom label other than an integer.

Try this:

import os
import tensorflow as tf
os.chdir(r'c:/users/user/Pictures')
from glob2 import glob

x_files = glob('inputs/*.jpg')
y_files = glob('targets/*.jpg')

files_ds = tf.data.Dataset.from_tensor_slices((x_files, y_files))

def process_img(file_path):
    img = tf.io.read_file(file_path)
    img = tf.image.decode_jpeg(img, channels=3)
    img = tf.image.convert_image_dtype(img, tf.float32)
    img = tf.image.resize(img, size=(28, 28))
    return img

files_ds = files_ds.map(lambda x, y: (process_img(x), process_img(y))).batch(1)

original, target = next(iter(files_ds))

<tf.Tensor: shape=(1, 28, 28, 3), dtype=float32, numpy=
array([[[[0.357423  , 0.3325731 , 0.20412168],
         [0.36274514, 0.21940777, 0.17623049],
         [0.34821934, 0.13921566, 0.06858743],
         ...,
         [0.25486213, 0.27446997, 0.2520612 ],
         [0.04925931, 0.26666668, 0.07619007],
         [0.48167226, 0.5287311 , 0.520888  ]]]

Then you won't need to pass an y to the fit() call. You will be able to use it as such:

model.fit(ds, epochs=5)