Four layer parameters of CNN are given in the table (figure attached). The first convolution layer has 1x9x9 input. After applying the 20 filters of size 4x4 how does the output 5x5x20?
can anybody help me to understand why the 5x5 output is?
I'm assuming that your input shapes have the last dimension as the channel dimension. Given that, the first input shape 1x9x9 should be 9x9x1 instead, as you cannot apply a 4x4 filter on an input with a shape 1x9. Applying 20 4x4 filters on a 9x9x1 input will result in the expected 5x5x20 output shape. It is probably just a typo.
Jandrik Lana Thanks, for your reply. But assuming 9x9x1 input instead of 1x9x9 with a 4x4 filter can not give 5x5 output. According to CNN
output=N-F+1=9-4+1=6.
Therefore, the output should be 6x6x20. However, 8x8x1 input with a 4x4 filter can give output=5x5.
Anil Kamboj I see now, thanks for pointing that out. The output should be 6x6x20, but the table says 5x5x20. I am also confused now about the table. The only thing I can think of is if the 4x4 filter has a dilation factor of 2 and a padding of 1, it will result in 5x5x20.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts