Keras image_dataset_from_directory not finding images

Question

I want to load the data from the directory where I have around 5000 images (type 'png'). But it returns me an error saying that there are no images when obviusly there are images. This code:

width=int(wb-wa)
height=int(hb-ha)
directory = '/content/drive/My Drive/Colab Notebooks/Hair/Images'
train_ds = tf.keras.preprocessing.image_dataset_from_directory(
    directory, labels=densitat, label_mode='int',
    color_mode='rgb', batch_size=32, image_size=(width, height), shuffle=True, seed=1,
    validation_split=0.2, subset='training', follow_links = False)

Returns:

ValueError: Expected the lengths of `labels` to match the number of files in the target directory. len(labels) is 5588 while we found 0 files in /content/drive/My Drive/Colab Notebooks/Hair/Images.

I can see the images: Colab view of the folder structure with the images

Where is the problem? I need to use this function to load data in batchs as i have a large dataset

Shrader

Answer 1

I have found the answer so I am posting in case it might help someone.

The problrem is the path, as I was using the path to the folder with the images whereas I should have used the directory (one folder above).

directory = '/content/drive/My Drive/Colab Notebooks/Hair'

Note that '/Hair' is the folder with my images.

Shrader

I have a similar problem using .tif files; however, moving to a directory above doesn't help. I used parent = ltifs_lib[0].parents[0].parents[0].parents[0].__str__() (where ltifs_lib[0] is the path to the first tif as given by .glob("**/*.tif") from pathlib) applying .parents[0] between 0 and 3 times. Doesn't help. Moreover, I provided a list with the labels. – Loudhailer

Answer 2

If the accepted solution above doesn't solve your problem, it could be because you are trying to load TIFF images with a .tif extension. It turns out the only allowed formats for image_dataset_from_directory are ('.bmp', '.gif', '.jpeg', '.jpg', '.png')

Quadruplet

So how can we load the .tif file from directory? – Converse

Answer 3

When you specify the labels parameter (as in the question), you need to follow the directory structure specified in the documentation (https://www.tensorflow.org/api_docs/python/tf/keras/utils/image_dataset_from_directory#expandable-1). You must give a parent directory containing as many directories as you are specifying in the labels parameter. And inside of each directory the images that correspond to that class.

When you specify a None as the labels parameter value, you can give the directory that contains all the images.

Hereditament