A single-channel feature map is about to go through one of two designs. Everything runs at stride 1 with no padding, every layer has exactly one filter, and only weights are counted (ignore biases).
| design | |
|---|---|
| Path A | one conv layer — weights |
| Path B | three stacked conv layers |
Path B's second layer reads the first layer's output, and the third reads the second's.
What does Path B produce, and how many weights does the whole of Path B hold?
Select all that apply.