One Layer Convolutional Network and Quizz

Hello all.
I have several questions, which I might guess is something I misunderstood, but I am getting confused in this quite often so I tried to post it the best way I can

In the video "One Layer Convolutional Network) by the end of the video Andrew shows a summary of the dimmensions of each component of the convolutional layer. The number of filters is given by n_c^{[l]}. Reported to the image of the toy example he gave by minute 4:36 he says we have 2 filters.

However, as I understood by Question 3 in my quizz (given an RGB image 256x256 and a convolutional layer with 128 filters with 7x7 shape) the correct answer implies that n_c^{[l-1]} is not the number of filters but the depth of the given image. Isn’t this an ambiguity in the equations? The same can be guessed for my Question 4 regarding the input and output equations of the same summary in the “One Layer Convolutional Network”: the n_c^{[l-1]} is refered to number of channels of the previous “image”, while n_c^{[l]} regards the number of filters on the conv layer, right?

The bottom line is: is not the parameter n_c^{[l]} being ambiguasly used to describe both the number of filters in n_c^{[l]} (as andrew tells in the video) AND the number of channels of the previous layer in n_c^{[l-1]}? to (which in fact suggestive, since “n_c” can easily be read as “number_(of)_channels”?

I hope I have not make the understanding too complicated :slight_smile:

Kind Regards.

Hi Ricardo_Gomes,

n_c^{[l]} is the depth (i.e. the number of channels) of the volume that is output by layer l. It is equal to the number of filters used in layer l. If layer l-1 wer an intermediate layer, the depth of its output volume would be equal to the number of filters used in layer l-1. Only in the first layer, in which the image is provided as an input volume, is the depth equal to the number of color channels (and has no relation to filters).
I hope this clarifies.