Course 4: Week 1: Reading: CNN Example *CORRECTION*

[user73]

hi, i’m trying to get my head around the figures given for the parameters in the cnn for the conv1 and conv2 layer

Here are screenshots for reference:

From my understanding
for the first conv layer,
the calculation should be (5 * 5 * 3 + 1) * 6.
i am not sure why the figure of 8 is used in the typo…
you have a 5 *5 *3 filter convolving on the input image.
there are 6 such filters, and each filter has a bias term, so (5 *5 *3+1) * 6

somewhat similarly in the 2nd conv layer
i get (5* 5* 6+1) * 16
the figure 8 in used in the typo in place of the 6.
from my understanding, you have a 5* 5* 6 filter convolving on the input image of size 14 * 14 * 6.
there are 16 such filters, and each filter has a bias term, so (5*5 *6+1) * 16.
i saw this topic raised in the forums but was directed here. there doesn’t seem to be an obvious way of searching for a topic so apologies if this has already been answered…
would appreciate some clarification on this!

thanks

[edwardyu]

Hi @user73 ,

Thanks for reporting the problem. You’re right, the #of channels in LeNet-5 Conv1 layer is 6 rather than 8. The issue is same as this.

[NN3]

Would be helpful to mention this explicitly in correction slide. Had to look around why my computed number of params are different.

[yzmlxxwch]

You solved my question. Thanks!

[KangarooBro]

i was wondering the same thing; great to spot the errors