Transfer_learning_with_MobileNet_v1 misunderstanding

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

Hi,
In the Transfer_learning_with_MobileNet assignment in paragraph 3.3 - Fine-tuning the Model written that I need to unfreez the base_model.

But Andrew says that we need to freeze the layers.
What this step of unfreezing do ?

2

Hi @Pavel_Grobov,

Let me quote from the lab:

First, unfreeze the base model by setting base_model.trainable=True , set a layer to fine-tune from, then re-freeze all the layers before it.

What do you think these steps are trying to do?

Does it finally freeze any layer?

Should some layers be unfreezed at the end so that those layers can be trained?

Raymond

3

Hi @rmwkwok ,

After reading this another time, now I understand this.

I need to unfreez all the layers and then refreez the layers from 120 to 155.

So yes finally it freeze the last layers.

Thank you!

4

Hi @Pavel_Grobov,

I am afraid you might want to check the following sentences again.

It said “re-freeze all the layers before it”. It said “re-freeze … before it”.

Cheers,
Raymond

5

Hi @rmwkwok ,

Ohh, I wanted to write that I need to retrain the layers between 120m and 155.