Keras ValueError: fit_generator yielding a list

Question

I'm learning how to work with Keras with TF backend for image recognition so I'm still not sure about what am I doing wrong here.

I'm trying to stack 2 models, one being VGG16, and the other one being a random one I made just to learn how to stack it. I want to classify an image among 5 classes.

The problem is in the last part, when I run fit_generator. Instead of yielding a valid tuple, it's yielding what it looks like to be a list. I've seen a lot of people getting similar problems, but in their cases, the output was None, so I'm not sure if the solution would be the same.

Parameters

nb_train_samples = 576
nb_validation_samples = 144
epochs = 30
batch_size = 12
img_width, img_height = 150, 150

Generators

train_datagen = ImageDataGenerator(
    rescale=1./255,
    rotation_range=50,
    width_shift_range=0.3,
    height_shift_range=0.3,
    shear_range=0.4,
    zoom_range=0.4,
    horizontal_flip=True)

test_datagen = ImageDataGenerator(rescale=1. / 255)

train_generator = train_datagen.flow_from_directory(
    train_data_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    class_mode=None,
    shuffle=False)

validation_generator = test_datagen.flow_from_directory(
    validation_data_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    class_mode=None,
    shuffle=False)

My model

input = Input(batch_shape=model.output_shape)
x = Flatten()(input)
x = Dense(256, activation='relu', name="new_block_1")(x)
x = Dropout(0.5)(x)
x = Dense(256, activation='relu', name="new_block_2")(x)
x = Dropout(0.5)(x)
x = Dense(5, activation='softmax', name="new_block_3")(x)
top_model = Model(input,x)

input = Input(shape=(img_width, img_height, 3))
x = model(input)
x = top_model(x)
final_model = Model(input, x)

final_model.compile(optimizer='rmsprop',
              loss='categorical_crossentropy', metrics=['accuracy'])

Fit and Error

final_model.fit_generator(
    train_generator,
    steps_per_epoch=nb_train_samples // batch_size,
    epochs=epochs,
    validation_data=validation_generator,
    validation_steps=nb_validation_samples // batch_size)

ValueError: output of generator should be a tuple `(x, y, sample_weight)` or `(x, y)`. Found: [[[[ 0.89411771  0.89019614  0.87450987]
   [ 0.89411771  0.89019614  0.87450987]
   [ 0.89411771  0.89019614  0.87450987]
   ...,

Update 1: as per @petezurich's tip, changed the activation function from 'sigmoid' to 'softmax'

This might not directly refer to your question, but is there a specific reason why you have a sigmoid on your last dense layer rather than a softmax? — petezurich, Jun 19 '17 at 19:19
no reason, actually, but might as well try changing it. I was just trying stuff out, so I might have copy-pasted it without noticing — htcoelho, Jun 19 '17 at 19:49
Softmax is the right choice in your case, since it gives you the probability of each of your 5 classes. Sigmoid would be correct for a binary classification. — petezurich, Jun 19 '17 at 19:51
I see. thank you for the tip :) anyway, I changed it to softmax, but sadly, the problem persists — htcoelho, Jun 19 '17 at 19:53
How do you get your labels? The easy way would be to put your images in subfolders for each class and set the `class_mode` in your generators to `True`. The generator than nicely derives the class labels from that. — petezurich, Jun 19 '17 at 20:06
there was no True, but there was a 'categorical' value, which actually worked. feel free to submit it as an answer, and I'll accept it. thank you very much :) — htcoelho, Jun 19 '17 at 20:13

score 1 · Accepted Answer · answered Jun 19 '17 at 20:16

1

Your model is missing labels for training.

Simply set the class_mode to categorical in your generators and put your images in subfolders for each class. The generator than nicely derives the class labels from that.

answered Jun 19 '17 at 20:16

petezurich

9,280
9
43
57

Keras ValueError: fit_generator yielding a list

1 Answers1