I have a set of images in a directory train_data = 'dataset/train' for handwritten Arabic letters, but it only contains the images in one folder and is not divided into subfolders in order for the ImageDataGenerator().flow_from_directory(train_data, class_mode='categorical', batch_size=64 function to understand the classes we want to classify.

When I first used Keras for CNN the images were divided in the training directory like this: dataset/training_set/cats and dataset/training_set/dogs and the same applies to the testing set directory, The output of ImageDataGenerator().flow_from_directory() gave me this line: Found 8000 images belonging to 2 classes.

This is how the images in the folder are named:

• id_1_label_1.png

• id_2_label_2.png

  And here is a picture so it is clearer: enter image description here Train data in my directory

The labels are part of the file name as shown above.

I tried running the following code to read the images:

train_datagen = ImageDataGenerator(rescale = 1./255,
                                   shear_range = 0.2,
                                   zoom_range = 0.2,
                                   horizontal_flip = True)
training_set = train_datagen.flow_from_directory('dataset/train',
                                                 target_size = (64, 64),
                                                 batch_size = 32,
                                                 class_mode = 'categorical')

But it gave me this output: Found 0 images belonging to 0 classes.

How can I load the data using Keras for CNN in order for ImageDataGenerator to understand how many classes we have?