how to use to_categorical when using ImageDataGenerator

Question

I am using Keras for classifying images (multiple classes) and I'm using ImageDataGenerator. It automatically finds all of classes, and it doesn't seem to write labels in any variable. I figured I need to use to_categorical to store my labels in matrix form, but I just don't know where to use it.

Here is a snippet of my code:

...
datagen = ImageDataGenerator(
    rotation_range=40,
    width_shift_range=0.2,
    height_shift_range=0.2,
    rescale=1./255,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True,
    fill_mode='nearest')

# generator for training
train_generator = datagen.flow_from_directory(
train_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode='categorical')

# generator for validation
val_generator = datagen.flow_from_directory(
val_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode='categorical')

# generator for testing
test_generator = datagen.flow_from_directory(
test_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode='categorical')

# train
model.fit_generator(
train_generator,
steps_per_epoch=nb_train_samples // batch_size,
epochs=epochs,
validation_data=val_generator,
validation_steps=nb_validation_samples // batch_size)

Generators just say "Found 442 images belonging to 5 classes." or smth like that. How can I use to_categorical on my labels?

Answer 1

It might be useful (even after two years) to also mention that if you want specific order for one-hot vectors, you can feed that through classes argument.
For example if you want "dog"=[1,0] and "cat"=[0,1], then explicitly set:
classes=["dog", "cat"].

Answer 2

Since you are passing class_mode='categorical' you dont have to manually convert the labels to one hot encoded vectors using to_categorical().

The Generator will return labels as categorical.