ValueError: Shape mismatch: The shape of labels (received (15,)) should equal the shape of logits except for the last dimension (received (5, 3))

Question

I am getting this error when trying to fit a model:

ValueError: Shape mismatch: The shape of labels (received (15,)) should equal the shape of logits except for the last dimension (received (5, 3)).

The code that's producing the error:

history = model.fit_generator(
  train_generator,
  epochs=10,
  validation_data=validation_generator)

This is the train_generator, validation generator is similar:

train_datagen = ImageDataGenerator(rescale=1./255)
train_generator = train_datagen.flow_from_directory(
        train_dir,
        target_size=(IMG_WIDTH, IMG_HEIGHT),
        batch_size=5)

I try to get the shapes:

for data_batch, labels_batch in train_generator:
    print('data batch shape:', data_batch.shape)
    print('labels batch shape:', labels_batch.shape)
    break

data batch shape: (5, 192, 192, 3) labels batch shape: (5, 3)

When I change the batch size, the shape of labels in the error changes accordingly (batch size of 3 gives an error with label shape (9,) for example, I have 3 classes). But my concern is that it is being done by the train_generator, is there anything I can do to fix this? Moreover, when I print the shapes from the train_generator, it seems right.

Here is the model, in case it is helpful:

model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
                        input_shape=(IMG_WIDTH, IMG_HEIGHT, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
for i in range(2):
  model.add(layers.Conv2D(128, (3, 3), activation='relu'))
  model.add(layers.MaxPooling2D((2, 2)))

model.add(layers.Flatten())
model.add(layers.Dense(3, activation='softmax'))


model.compile(optimizer='adam',
          loss='sparse_categorical_crossentropy',
          metrics=['accuracy'])

Thanks!

Edit - Complete code:

The directory contains two folders - train and validation and each of them has three subfolders with the images of the corresponding classes.

try:
 %tensorflow_version 2.x # enable TF 2.x in Colab
except Exception:
  pass

from tensorflow.keras import datasets, layers, models

IMG_WIDTH = 192
IMG_HEIGHT = 192
train_dir = 'train'
validation_dir = 'validation'

from google.colab import drive
drive.mount('/content/drive')

import os
os.chdir("drive/My Drive/colab")

from tensorflow.keras.preprocessing.image import ImageDataGenerator

train_datagen = ImageDataGenerator(rescale=1./255)
train_generator = train_datagen.flow_from_directory(
        train_dir,
        target_size=(IMG_WIDTH, IMG_HEIGHT),
        batch_size=5)

validation_datagen = ImageDataGenerator(rescale=1./255)
validation_generator = validation_datagen.flow_from_directory(
        validation_dir,
        target_size=(IMG_WIDTH, IMG_HEIGHT),
        batch_size=5)

model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
                        input_shape=(IMG_WIDTH, IMG_HEIGHT, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
for i in range(2):
  model.add(layers.Conv2D(128, (3, 3), activation='relu'))
  model.add(layers.MaxPooling2D((2, 2)))

model.add(layers.Flatten())
model.add(layers.Dense(3, activation='softmax'))

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

history = model.fit_generator(
      train_generator,
      epochs=10,
      validation_data=validation_generator)

Thanks!

Answer 1

The difference between sparse_categorical_crossentropy and categorical_crossentropy is whether your targets are one-hot encoded.

The shape of label batch is (5,3), which means it has been one-hot encoded. So you should use categorical_crossentropy loss function.

model.compile(optimizer='adam',
              loss='categorical_crossentropy',
              metrics=['accuracy'])

Answer 2

I think it is related to your loss function just try to use "categorical_crossentropy" instead of "sparse..."