I am training a food image classification deep learning model using TensorFlow/Keras, but I am encountering a strange issue where some epochs are skipped, and the training logs show 0 accuracy and 0 loss for those epochs. Here’s what I have observed:

•   Some epochs report 0 accuracy and 0 loss, while others seem to train correctly.
•   The data generators are set up correctly, and the number of images in the training and validation datasets are as expected.

Environment:

•   TensorFlow version: 2.x
•   Python version: 3.8
•   Dataset: Food-101 (with 101 classes)

What I Have Tried:

1.  Verified Data Directory Structure: Ensured that the directory structure is correct and contains the expected number of images.
2.  Simplified Model: Used a simpler model to rule out complexity-related issues.
3.  Removed Callbacks: Removed all callbacks to check if they were causing the problem.
4.  Calculated Steps per Epoch: Ensured that steps_per_epoch and validation_steps are correctly calculated based on the number of samples and batch size.
5.  Monitored Batch Processing: Added debugging callbacks to monitor batch processing and ensure consistency.

My code:

# Create ImageDataGenerators for data augmentation
train_datagen = ImageDataGenerator(
    rescale=1./255,
    rotation_range=30,
    width_shift_range=0.2,
    height_shift_range=0.2,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True,
    fill_mode='nearest',
    validation_split=0.2
)

# Create training and validation data generators
train_generator = train_datagen.flow_from_directory(
    data_dir,
    target_size=(150, 150),  # Use a smaller target size for faster debugging
    batch_size=32,
    class_mode='categorical',
    subset='training'
)

validation_generator = train_datagen.flow_from_directory(
    data_dir,
    target_size=(150, 150),
    batch_size=32,
    class_mode='categorical',
    subset='validation'
)

# Verify the number of samples
print(f"Number of training samples: {train_generator.samples}")
print(f"Number of validation samples: {validation_generator.samples}")

# Build a simpler model for debugging
model = Sequential([
    Flatten(input_shape=(150, 150, 3)),
    Dense(128, activation='relu'),
    Dropout(0.5),
    Dense(101, activation='softmax')  # Adjust the number of classes if necessary
])

# Compile the model
model.compile(
    optimizer=Adam(learning_rate=0.001),
    loss='categorical_crossentropy',
    metrics=['accuracy']
)

# Calculate steps per epoch and validation steps
steps_per_epoch = train_generator.samples // train_generator.batch_size
validation_steps = validation_generator.samples // validation_generator.batch_size
print(f"Steps per epoch: {steps_per_epoch}")
print(f"Validation steps: {validation_steps}")

# Train the model without callbacks

history = model.fit(
    train_generator,
    steps_per_epoch=steps_per_epoch,
    validation_data=validation_generator,
    validation_steps=validation_steps,
    epochs=50
)

Debugging Steps Taken:

1.  Verified the directory structure and number of images in each class.
2.  Simplified the model to rule out complexity-related issues.
3.  Removed all callbacks to check if they were causing the problem.
4.  Ensured the steps_per_epoch and validation_steps are correctly calculated.

Please help with this isse , not sure what to try next