yes because the Colab give us the instruction:
# Average pooling 2D layer.
# Used instead of GlobalAveragePooling2D for compatibility with the current
# version of the grader. Please do not modify. (Date: 08/19/2021)
I tried with the GlobalAveragePooling2D so my model summary was exactly the same as in the Colab, but still the same error message in exercise 7.
You are right there has been an update. Still though when I run the colab with the update line I dont get stuck. Your error says the input shape is not right, I think there should be something wrong with the inputs to the model somehow. I would check that.
I have similar problem about my validation result. I only get result like this:
Number of predictions where iou > threshold(0.5): 137 (27.4%)
even though my training error is pretty low. My model is as below:
which looks the same as the expected output.
In the bounding box regression layer, I have:
bounding_box_regression_output = tf.keras.layers.Dense(4, name=‘bounding_box’)(last_dense_layer)
And for compilation, my setting is:
optimizer = tf.keras.optimizers.SGD(momentum=0.9)
model.compile(optimizer=optimizer, loss = {‘bounding_box’: ‘mse’})
Is there anything wrong with these settings?
Never mind. After I restart the notebook and rerun everything, I get a result that can pass the 45% of >50 iou.