Search code examples
tensorflowmachine-learningkerasneural-network

flow_from_directory fitting of model yields ValueError: Input 0 is incompatible with layer model

I have the following model:

enter image description here

I tried to fit the model using ImageDataGenerator with flow*_*from_directory and fit_generator, however I get the following error:

ValueError: Input 0 is incompatible with layer model: expected shape=(None, 256, 256, 1), found shape=(None, 400, 400, 1)

I used the correct target_size, so I don't know why the error appears. My code is the following:

model_merged.compile(loss='categorical_crossentropy',
          optimizer="adam",
          metrics=['acc'])
train_datagen =ImageDataGenerator(rescale=1./255, validation_split=0.25)

#training data
train_generator = train_datagen.flow_from_directory(
    '/kaggle/working/images/',  # Source directory
    target_size=(256, 256),  # Resizes images
    batch_size=batch_size,
    class_mode='categorical',subset = 'training')

epochs = epochs
#Testing data
validation_generator = train_datagen.flow_from_directory(
    '/kaggle/working/images/',
    target_size=(256, 256),
    batch_size=batch_size,
    class_mode='categorical',
    subset='validation') # set as validation data
#Model fitting for a number of epochs
history = model_merged.fit_generator(
  train_generator,
  steps_per_epoch=steps_train,
  epochs=epochs,
  validation_data = validation_generator,
  validation_steps = steps_val,
  verbose=1)

Update

batch_size = 32
epochs = 32
steps_train = 18
steps_val = 3
img_height = 256
img_width = 256
data_dir='/kaggle/working/images/'
model_merged.compile(loss='categorical_crossentropy',
          optimizer="adam",
          metrics=['acc'])
train_ds = tf.keras.utils.image_dataset_from_directory(
  data_dir,
  validation_split=0.2,
  subset="training",
  seed=123,
  image_size=(img_height, img_width),
  batch_size=batch_size)

val_ds = tf.keras.utils.image_dataset_from_directory(
  data_dir,
  validation_split=0.2,
  subset="validation",
  seed=123,
  image_size=(img_height, img_width),
  batch_size=batch_size)

# scale pixel value between 0 and 1
normalization_layer = tf.keras.layers.Rescaling(1./255)
reshape_layer = Reshape((-1,256,256))
resize_layer = Resizing(1,256)
permute_layer = Permute((2,3,1))

train_ds = train_ds.map(lambda x, y: (normalization_layer(x), y))
val_ds = val_ds.map(lambda x, y: (normalization_layer(x), y))
train_ds = train_ds.map(lambda x, y: (reshape_layer(x), y))
val_ds = val_ds.map(lambda x, y: (reshape_layer(x), y))
train_ds = train_ds.map(lambda x, y: (resize_layer(x), y))
val_ds = val_ds.map(lambda x, y: (resize_layer(x), y))
train_ds = train_ds.map(lambda x, y: (permute_layer(x), y))
val_ds = val_ds.map(lambda x, y: (permute_layer(x), y))


AUTOTUNE = tf.data.AUTOTUNE

train_ds = train_ds.cache().prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)

history = model_merged.fit(
  train_ds,
  steps_per_epoch=steps_train,
  epochs=epochs,
  validation_data = val_ds,
  validation_steps = steps_val,
  verbose=1)

The error is the same as the one above. I added some new layers in order to change the input layer from (None,256,256,3) to (None,256,256,1), as this was the initial error, but it still doesn't work. I am not sure where the error comes from because the dimensions of the train and validation datasets are correct now.

Update 2 new model I removed the concatenation layer from the merged model, since I want the output of model A to be passed to the input of model B, however, even with the new merged model the error still appears.

Solution

  • Are you open to use the more modern https://www.tensorflow.org/api_docs/python/tf/keras/utils/image_dataset_from_directory

    if yes this is a working example close to your code

    epochs = 32
steps_train = 18
steps_val = 3
img_height = 256
img_width = 256
data_dir='/kaggle/working/images/'
model_merged.compile(loss='categorical_crossentropy',
optimizer="adam",
metrics=['acc'])
train_ds = tf.keras.utils.image_dataset_from_directory(
data_dir,
validation_split=0.25,
subset="training",
seed=123,
image_size=(img_height, img_width),
label_mode="categorical",
batch_size=batch_size)

val_ds = tf.keras.utils.image_dataset_from_directory(
data_dir,
validation_split=0.25,
subset="validation",
seed=123,
image_size=(img_height, img_width),
label_mode="categorical",
batch_size=batch_size)

# scale pixel value between 0 and 1
normalization_layer = tf.keras.layers.Rescaling(1./255)
reshape_layer = Reshape((-1,256,256))
resize_layer = Resizing(1,256)
permute_layer = Permute((2,3,1))

train_ds = train_ds.map(lambda x, y: (normalization_layer(x), y))
val_ds = val_ds.map(lambda x, y: (normalization_layer(x), y))
train_ds = train_ds.map(lambda x, y: (reshape_layer(x), y))
val_ds = val_ds.map(lambda x, y: (reshape_layer(x), y))
train_ds = train_ds.map(lambda x, y: (resize_layer(x), y))
val_ds = val_ds.map(lambda x, y: (resize_layer(x), y))
train_ds = train_ds.map(lambda x, y: (permute_layer(x), y))
val_ds = val_ds.map(lambda x, y: (permute_layer(x), y))


AUTOTUNE = tf.data.AUTOTUNE

train_ds = train_ds.cache().prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)

history = model_merged.fit(
train_ds,
steps_per_epoch=steps_train,
epochs=epochs,
validation_data = val_ds,
validation_steps = steps_val,
verbose=1)