tensorflow: tensorflow/core/grappler/optimizers/meta_optimizer.cc:954] layout failed: INVALID_ARGUMENT: Size of values 0 does not match size of permutation 4 @ fanin shape inmodel_6/dropout_12/dropout/SelectV2-2-TransposeNHWCToNCHW-LayoutOptimizer

Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

No

Source

source

TensorFlow version

2.12.0

Custom code

Yes

OS platform and distribution

Kaggle

Mobile device

No response

Python version

3.10.12

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current behavior?

I am trying to train model below

import tensorflow as tf
from tensorflow.keras.models import Model
from tensorflow.keras.layers import *

# Set the input shape of the images (adjust based on the input image size)
input_shape = (128, 128, 3)  # Adjust based on the input image size

# Set the number of segmentation classes 
n_classes = 1  # Number of segmentation classes

# Define the model architecture
inputs = Input(shape=input_shape)  # Define the input layer with the specified input shape

# Encoder
conv1 = Conv2D(64, (3, 3), activation='relu', padding='same')(inputs)  # First convolutional layer with 64 filters
conv1 = BatchNormalization()(conv1)  # Apply batch normalization to normalize the activations
conv1 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv1)  # Second convolutional layer with 64 filters
conv1 = BatchNormalization()(conv1)  # Apply batch normalization to normalize the activations
pool1 = MaxPooling2D((2, 2))(conv1)  # Max pooling layer with a pool size of (2, 2)

conv2 = Conv2D(128, (3, 3), activation='relu', padding='same')(pool1)  # Convolutional layer with 128 filters
conv2 = BatchNormalization()(conv2)  # Apply batch normalization to normalize the activations
conv2 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv2)  # Convolutional layer with 128 filters
conv2 = BatchNormalization()(conv2)  # Apply batch normalization to normalize the activations
pool2 = MaxPooling2D((2, 2))(conv2)  # Max pooling layer with a pool size of (2, 2)

conv3 = Conv2D(256, (3, 3), activation='relu', padding='same')(pool2)  # Convolutional layer with 256 filters
conv3 = BatchNormalization()(conv3)  # Apply batch normalization to normalize the activations
conv3 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv3)  # Convolutional layer with 256 filters
conv3 = BatchNormalization()(conv3)  # Apply batch normalization to normalize the activations
pool3 = MaxPooling2D((2, 2))(conv3)  # Max pooling layer with a pool size of (2, 2)

conv4 = Conv2D(512, (3, 3), activation='relu', padding='same')(pool3)  # Convolutional layer with 512 filters
conv4 = BatchNormalization()(conv4)  # Apply batch normalization to normalize the activations
conv4 = Conv2D(512, (3, 3), activation='relu', padding='same')(conv4)  # Convolutional layer with 512 filters
conv4 = BatchNormalization()(conv4)  # Apply batch normalization to normalize the activations
drop4 = Dropout(0.5)(conv4)  # Apply dropout regularization with a rate of 0.5
pool4 = MaxPooling2D((2, 2))(drop4)  # Max pooling layer with a pool size of (2, 2)

conv5 = Conv2D(1024, (3, 3), activation='relu', padding='same')(pool4)  # Convolutional layer with 1024 filters
conv5 = BatchNormalization()(conv5)  # Apply batch normalization to normalize the activations
conv5 = Conv2D(1024, (3, 3), activation='relu', padding='same')(conv5)  # Convolutional layer with 1024 filters
conv5 = BatchNormalization()(conv5)  # Apply batch normalization to normalize the activations
drop5 = Dropout(0.5)(conv5)  # Apply dropout regularization with a rate of 0.5

# Decoder
up6 = concatenate([UpSampling2D((2, 2))(drop5), conv4], axis=-1)  # Upsampling layer with a scale factor of (2, 2)
conv6 = Conv2D(512, (3, 3), activation='relu', padding='same')(up6)  # Convolutional layer with 512 filters
conv6 = BatchNormalization()(conv6)  # Apply batch normalization to normalize the activations
conv6 = Conv2D(512, (3, 3), activation='relu', padding='same')(conv6)  # Convolutional layer with 512 filters
conv6 = BatchNormalization()(conv6)  # Apply batch normalization to normalize the activations

up7 = concatenate([UpSampling2D((2, 2))(conv6), conv3], axis=-1)  # Upsampling layer with a scale factor of (2, 2)
conv7 = Conv2D(256, (3, 3), activation='relu', padding='same')(up7)  # Convolutional layer with 256 filters
conv7 = BatchNormalization()(conv7)  # Apply batch normalization to normalize the activations
conv7 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv7)  # Convolutional layer with 256 filters
conv7 = BatchNormalization()(conv7)  # Apply batch normalization to normalize the activations

up8 = concatenate([UpSampling2D((2, 2))(conv7), conv2], axis=-1)  # Upsampling layer with a scale factor of (2, 2)
conv8 = Conv2D(128, (3, 3), activation='relu', padding='same')(up8)  # Convolutional layer with 128 filters
conv8 = BatchNormalization()(conv8)  # Apply batch normalization to normalize the activations
conv8 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv8)  # Convolutional layer with 128 filters
conv8 = BatchNormalization()(conv8)  # Apply batch normalization to normalize the activations

up9 = concatenate([UpSampling2D((2, 2))(conv8), conv1], axis=-1)  # Upsampling layer with a scale factor of (2, 2)
conv9 = Conv2D(64, (3, 3), activation='relu', padding='same')(up9)  # Convolutional layer with 64 filters
conv9 = BatchNormalization()(conv9)  # Apply batch normalization to normalize the activations
conv9 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv9)  # Convolutional layer with 64 filters
conv9 = BatchNormalization()(conv9)  # Apply batch normalization to normalize the activations

outputs = Conv2D(n_classes, (1, 1), activation='softmax')(conv9)  # Convolutional layer for output

# Create the model
model = Model(inputs=inputs, outputs=outputs)

# Print the model summary
model.summary()


# Set the optimizer for the model
optimizer = tf.keras.optimizers.Adam(lr=1e-4)

# Compile the model with loss function and metrics
model.compile(optimizer=optimizer, loss='binary_crossentropy', metrics=['accuracy'])

images_path = '/kaggle/input/coco-2014-dataset-for-yolov3/coco2014/images/val2014'
masks_path = '/kaggle/working/mask_val_2014'
batch_size = 8

val_generator = CustomDataGenerator(images_path, masks_path, batch_size)

Also to ensure that all the input and output have proper shape I run the code below

def print_preprocessed_image_shapes(model, generator):
    """
    Print the shapes of preprocessed images generated by the provided model and generator.

    Args:
        model (tf.keras.Model): The trained model.
        generator (CustomDataGenerator): Instance of the CustomDataGenerator class.
    """
    for i in range(len(generator)):
        # Get a batch of preprocessed images from the generator
        batch_images, batch_masks = generator[i]

        # Print the shapes of the preprocessed images
        for image in batch_images:
            print(f"Shape of preprocessed image: {image.shape}")
        for mask in batch_maskss:

             print(f"Shape of preprocessed image: {mask.shape}")
            
# Print the shapes of preprocessed images
print_preprocessed_image_shapes(model, val_generator)

As a result of this error, the model was unable to undergo the training process.

Standalone code to reproduce the issue

# Fit the model with the training generator
train_steps =  len(os.listdir( "/kaggle/working/mask_train_2014/"))/batch_size
model.fit(train_generator,validation_data = val_generator, steps_per_epoch = train_steps , epochs=20)

Relevant log output

tensorflow/core/grappler/optimizers/meta_optimizer.cc:954] layout failed: INVALID_ARGUMENT: Size of values 0 does not match size of permutation 4 @ fanin shape inmodel_5/dropout_10/dropout/SelectV2-2-TransposeNHWCToNCHW-LayoutOptimizer