CODEWITHSUNDEEP
pythontensorflowkeras
Nikit Parakh
Nikit Parakh

Reputation: 75

Tensorflow: Logits and Label must be the same size

I am currently attempting a project in Google/Udacity's Tensorflow Course using a dataset acquired as follows:

_URL = "https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz"

zip_file = tf.keras.utils.get_file(origin=_URL,
                                   fname="flower_photos.tgz",
                                   extract=True)

Unfortunately, I ran into the following error:

InvalidArgumentError:  logits and labels must have the same first dimension, got logits shape [100,5] and labels shape [500]
     [[node sparse_categorical_crossentropy/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits (defined at <ipython-input-43-02964d57939c>:8) ]] [Op:__inference_test_function_3591]

I looked at other posts, but it still seemed a bit tricky to figure out. My initial thought is that I might be using the incorrect loss function.

Here is the code running into problems:

image_gen = ImageDataGenerator(rescale = 1./255, horizontal_flip=True, zoom_range=0.5, rotation_range=45, width_shift_range=0.15, height_shift_range=0.15)

train_data_gen = image_gen.flow_from_directory(batch_size=BATCH_SIZE, directory = train_dir, shuffle=True, target_size=(IMG_SHAPE,IMG_SHAPE),class_mode='binary')

image_gen = ImageDataGenerator(rescale = 1./255)

val_data_gen = image_gen.flow_from_directory(batch_size=BATCH_SIZE, directory = val_dir, shuffle=True, target_size=(IMG_SHAPE,IMG_SHAPE))


model = tf.keras.models.Sequential([
                                    tf.keras.layers.Conv2D(16, (3,3), activation='relu', input_shape=(150,150,3)),
                                    tf.keras.layers.MaxPooling2D(2,2),
                                    tf.keras.layers.Conv2D(32, (3,3), activation='relu'),
                                    tf.keras.layers.MaxPooling2D(2,2),
                                    tf.keras.layers.Conv2D(64, (3,3), activation='relu'),
                                    tf.keras.layers.MaxPooling2D(2,2),
                                    tf.keras.layers.Dropout(0.5),
                                    tf.keras.layers.Flatten(),
                                    tf.keras.layers.Dense(512, activation='relu'),
                                    tf.keras.layers.Dense(5),
                                    
])

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])
model.summary()

The batch size is 100 and input dimension is 150,150 The summary is as follows: Model: "sequential_4"

Layer (type) Output Shape Param #

conv2d_12 (Conv2D) (None, 148, 148, 16) 448

max_pooling2d_12 (MaxPooling (None, 74, 74, 16) 0

conv2d_13 (Conv2D) (None, 72, 72, 32) 4640

max_pooling2d_13 (MaxPooling (None, 36, 36, 32) 0

conv2d_14 (Conv2D) (None, 34, 34, 64) 18496

max_pooling2d_14 (MaxPooling (None, 17, 17, 64) 0

dropout_4 (Dropout) (None, 17, 17, 64) 0

flatten_4 (Flatten) (None, 18496) 0

dense_8 (Dense) (None, 512) 9470464

dense_9 (Dense) (None, 5) 2565

Total params: 9,496,613 Trainable params: 9,496,613 Non-trainable params: 0

Any thoughts on what may be wrong?

Upvotes: 1

Views: 5976

Answers (2)

Akshay Malik
Akshay Malik

Reputation: 1

In the generator I updated the class_mode as 'sparse' and it worked fine.

train_data_gen = image_gen.flow_from_directory(train_dir, target_size = (IMG_SHAPE, IMG_SHAPE), batch_size = batch_size, class_mode = 'sparse')

Upvotes: 0

Marco Cerliani
Marco Cerliani

Reputation: 22021

pay attention to class_mode in your generator

'int': means that the labels are encoded as integers (e.g. for sparse_categorical_crossentropy loss). 'categorical' means that the labels are encoded as a categorical vector (e.g. for categorical_crossentropy loss). 'binary' means that the labels (there can be only 2) are encoded as float32 scalars with values 0 or 1 (e.g. for binary_crossentropy). None (no labels).

it seems you need 'int' instead of 'binary' for both train and validation generator

Upvotes: 4

Related Questions