Classic Network: LeNet 5

General clarification on Convolution layer, but citing the example from LeNet 5

  1. Input is 32*32*1
  2. First conv layer has 6 3*3 filters, with s=1 and p=0. So output of this is 28*28*6
  3. First Avg Pooling layer has s=2 and f=2. so output is 14*14*6 volume
  4. In the second layer in the video professor says that the filters are 5*5. But my question is, since the input to the second layer is a 14*14*6 volume, wouldnt the filter in this layer also be 5*5*6

Yes, my guess is the only reason Prof Ng does not explicitly mention that is that it’s not a choice, right? You have a choice of f at each layer and you have a choice of how many filters you have, but you don’t have a choice about the input channels: the filters have to match the input channels.

Also note that I edited your post to use the “{}” formatting tool on all the filter shapes, so that we can actually read them. The point is that “*” means “italics” in markdown, so you have to make sure that any mathematical expressions get formatted “as is”, not as markdown.

Thank you for the clarification, i just wanted to make sure that I was right about the dimension,

And thank you for explaining about my missing *, I saw that after I posted

Another way is to put a backslash before the asterisk. e.g. 3 \* 5 = 15 gives 3 * 5 = 15.