Section 5 - transfer learning feature extraction input shape error

[ngcheeyuan]

Tensorflow Version 2.7. I'm faced with an input shape warning when executing the code provided. Please assist. Thank you.

**** below is the code and error ******

input_shape = (224, 224, 3)
base_model = tf.keras.applications.EfficientNetB0(include_top=False)
base_model.trainable = False

### Create input layer
inputs = layers.Input(shape=input_shape, name="input_layer")

### Add in data augmentation Sequential model as a layer
x = data_augmentation(inputs)

### Give base_model inputs (after augmentation) and don't train it
x = base_model(x, training=False)

### Pool output features of base model
x = layers.GlobalAveragePooling2D(name="global_average_pooling_layer")(x)

### Put a dense layer on as the output
outputs = layers.Dense(10, activation="softmax", name="output_layer")(x)

### Make a model with inputs and outputs
model_1 = keras.Model(inputs, outputs)

### Compile the model
model_1.compile(loss="categorical_crossentropy",
              optimizer=tf.keras.optimizers.Adam(),
              metrics=["accuracy"])

### Fit the model
history_1_percent = model_1.fit(train_data_1_percent,
                    epochs=5,
                    steps_per_epoch=len(train_data_1_percent),
                    validation_data=test_data_10_percent,
                    validation_steps=int(0.25* len(test_data_10_percent)),
                    # Track model training logs
                    callbacks=[create_tensorboard_callback("transfer_learning_part_2", "1_percent_data_aug")])

WARNING:tensorflow:Model was constructed with shape (287, 512, 3) for input KerasTensor(type_spec=TensorSpec(shape=(287, 512, 3), dtype=tf.uint8, name='random_flip_5_input'), name='random_flip_5_input', description="created by layer 'random_flip_5_input'"), but it was called on an input with incompatible shape (None, 224, 224, 3).
Saving TensorBoard log files to: transfer_learning_part_2/1_percent_data_aug/20220111-164520
Epoch 1/5
WARNING:tensorflow:Model was constructed with shape (287, 512, 3) for input KerasTensor(type_spec=TensorSpec(shape=(287, 512, 3), dtype=tf.uint8, name='random_flip_5_input'), name='random_flip_5_input', description="created by layer 'random_flip_5_input'"), but it was called on an input with incompatible shape (None, 224, 224, 3).
/usr/local/lib/python3.7/dist-packages/keras/engine/functional.py:1410: CustomMaskWarning: Custom mask layers require a config and must override get_config. When loading, the custom mask layer must be passed to the custom_objects argument.
  layer_config = serialize_layer_fn(layer)
WARNING:tensorflow:Model was constructed with shape (287, 512, 3) for input KerasTensor(type_spec=TensorSpec(shape=(287, 512, 3), dtype=tf.uint8, name='random_flip_5_input'), name='random_flip_5_input', description="created by layer 'random_flip_5_input'"), but it was called on an input with incompatible shape (None, 224, 224, 3).
3/3 [==============================] - ETA: 0s - loss: 2.4418 - accuracy: 0.0714WARNING:tensorflow:Model was constructed with shape (287, 512, 3) for input KerasTensor(type_spec=TensorSpec(shape=(287, 512, 3), dtype=tf.uint8, name='random_flip_5_input'), name='random_flip_5_input', description="created by layer 'random_flip_5_input'"), but it was called on an input with incompatible shape (None, 224, 224, 3).
3/3 [==============================] - 52s 22s/step - loss: 2.4418 - accuracy: 0.0714 - val_loss: 2.2500 - val_accuracy: 0.1924
Epoch 2/5
3/3 [==============================] - 45s 21s/step - loss: 2.1990 - accuracy: 0.2000 - val_loss: 2.1427 - val_accuracy: 0.2582
Epoch 3/5
3/3 [==============================] - 45s 22s/step - loss: 1.9974 - accuracy: 0.3000 - val_loss: 2.0428 - val_accuracy: 0.3125
Epoch 4/5
3/3 [==============================] - 45s 21s/step - loss: 1.8233 - accuracy: 0.4571 - val_loss: 1.9495 - val_accuracy: 0.3701
Epoch 5/5
3/3 [==============================] - 35s 17s/step - loss: 1.7141 - accuracy: 0.5714 - val_loss: 1.8388 - val_accuracy: 0.4523

[mrdbourke]

Hi @ngcheeyuan,

Looks like your model was training correctly, is this correct?

But the warning is still there - I feel like this may be an error (or an unnecessary warning).

Can you confirm the shapes of your input data & data augmentation layers? (I think the error may be within one of the data augmentation layers, because of random_flip_5_input, though this may be because of TensorFlow aug layers rather than your code)

How do these look?

Is your model still able to make sound predictions?

[ngcheeyuan]

Hi.

I've reran the code again and it seems to work.
Unable to replicate the error.

Currently working through the NLP portion of the course and it's been really awesome thus far.

Thanks!