#vgg-blocks

Question 1: Single choice

VGG-11 is built with arch = ((1, 64), (1, 128), (2, 256), (2, 512), (2, 512)), where each pair gives the number of $3 \times 3$ convolutions and the output channels of one VGG block, followed by three fully connected layers. Printing the output shape of each top-level module shows the five blocks and the three FC layers (plus flatten, ReLU and dropout), only eight entries with weights, although VGG-11 has 11 weight layers. Where are the other three?

They are the max-pooling layers between the blocks, which count as layers1
They are ReLU activations, which VGG counts among its weight layers2
Inside the last three blocks, each of which stacks two $3 \times 3$ convolutions3
Inside the first three blocks, each of which stacks two $3 \times 3$ convolutions4

Timed test