CODEWITHSUNDEEP
pythonpython-3.xkeras
Joe
Joe

Reputation: 279

Determining output of model.predict in relation to classess

I'm trying to figure out how to determine the class from a one hot encoded vector that is returned from keras. The problem is I use the flow_from_directory with ImageDataGenerator to train my network, leading to keras automatically converting the folder structure into the class vector to my knowledge, how do I go about resolving this?

Here's how my training is laid out in code:

    checkpoint = [ModelCheckpoint(
        'model.checkpoint.hdf5',
        period=1
   )]

    train_datagen = ImageDataGenerator(
        rotation_range=30)

    test_datagen = ImageDataGenerator(
        rotation_range=30)

    train_generator = train_datagen.flow_from_directory(
        '/train/',
        target_size=(x, y),
        batch_size=batch_size,
        class_mode='categorical')
    test_generator = test_datagen.flow_from_directory(
        '/test/',
        target_size=(x, y),
        batch_size=batch_size,
        class_mode='categorical')

    self.discriminator.fit_generator(
        train_generator,
        steps_per_epoch=10000,
        epochs=epochs,
        validation_data=test_generator,
        validation_steps=10,
        callbacks=checkpoint
    )

My folder structure is like so:

root/
    train/
        cow/
            0.jpg
            1.jpg
        pig/
            0.jpg
            1.jpg
    test/
        cow/
            0.jpg
            1.jpg
        pig/
            0.jpg
            1.jpg

Upvotes: 0

Views: 40

Answers (2)

Manoj Mohan
Manoj Mohan

Reputation: 6044

print(train_generator.class_indices)
{'cow': 0, 'pig': 1}

From the documentation, "The dictionary containing the mapping from class names to class indices can be obtained via the attribute class_indices."

https://keras.io/preprocessing/image/#flow_from_directory

Upvotes: 1

Zrufy
Zrufy

Reputation: 453

The prediction in keras refers to your own structure of the folder. if train is composed of cow and pig. Then cow = 0 and pig = 1. So if the predict returns 0 is cow otherwise it's pig.

Upvotes: 1

Related Questions