Epochs that are multiples of 2 are finishing very quickly. I don’t think my model is being trained. I’m using tensorflow keras to train my model. I am using colab environment.

Here is the results of training: Model training result screenshot

I am using the ResNet architecture. My batch size is 32. However, inexplicably, the epochs that are multiples of 2 finish quickly, and the accuracy of my model increases in an inexplicable way. I have 4 classes. 4027 images for training, 1150 images for validating, 577 images for testing. What could be the reason for this?

Here is what my code looks like:

data processing part:

train_dir = '/splitted-dataset/train'
validation_dir = '/splitted-dataset/val'
test_dir = '/splitted-dataset/test'

train_datagen = ImageDataGenerator(
    rescale=1./255,
    rotation_range=15,
    width_shift_range=0.1,
    height_shift_range=0.1,
    horizontal_flip=True
)

validation_datagen = ImageDataGenerator(rescale=1./255)
test_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory(
    train_dir,
    target_size=(256, 256),
    batch_size=32,
    class_mode='categorical'
)

validation_generator = validation_datagen.flow_from_directory(
    validation_dir,
    target_size=(256, 256),
    batch_size=32,
    class_mode='categorical'
)

test_generator = test_datagen.flow_from_directory(
    test_dir,
    target_size=(256, 256),
    batch_size=32,
    class_mode='categorical'
)

model structure part :

input_tensor = Input(shape=(256, 256, 3))
resnet_model = ResNet50(weights='imagenet', include_top=False, input_tensor=input_tensor)

x = resnet_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(256, activation='relu')(x)
x = BatchNormalization()(x)
x = Dropout(0.5)(x)
output_tensor = Dense(train_generator.num_classes, activation='softmax')(x)

model = Model(inputs=input_tensor, outputs=output_tensor)

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

model training part :

history = model.fit(
    train_generator,
    steps_per_epoch=train_generator.samples // train_generator.batch_size,
    epochs=10,
    validation_data=validation_generator,
)

test_loss, test_acc = model.evaluate(test_generator, verbose=2)
print(f"Test Accuracy: {test_acc:.2f}")

I tried different batch sizes and L2 regularization, but I couldn’t get useful results.