C4 W4 U-Net assignment diagram shows 1024 filters, but code goes to 512 only

In the U-net assignment, the diagrams show that we start with 64 filters, and end with 1024 at the bottom, and then go back up to 64 filters.

But in the code we start with 32 and have 512 at the bottom. Why the difference? Is it just to make it computationally faster and save on resources?

Edit: Additionally, there seems to be an extra convolution that isn’t even in the diagram, it is ‘conv9’ in the model code. I am unsure as to why it’s there. In the assignment, it says there are 23 convolutions in total. But ‘conv9’ bumps it to 24.

Figure 2 is incorrect. That’s also where “23 convolutions” is mentioned.
Figure 3 is correct.