What does the number 16 (no. of outputs) represent in this Convo layer?
layers.Conv2D(16, 3, padding='same', activation='relu'),
$\begingroup$
$\endgroup$
1
-
$\begingroup$ Hi @ShushilKgadka, welcome to the site. Here you have some context on the concept of "channels" in a convolutional layer. $\endgroup$– noeJun 17 at 22:27
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
It means that you have 16 filters of 3x3 kernels. If you check the number of trainable parameters, you have : 16 filters x 9 parameters per kernel x 3 channels (for RGB images) + 16 biaises (one per filter) = 448 trainable parameters
-
$\begingroup$ So the software knows that both dimensions are $3$ pixels if you only specify one number? I feel like I’ve always specified both dimensions. $\endgroup$– DaveAug 18, 2022 at 21:35
-
$\begingroup$ If you mean the size of the kernel yes , it does understand its a 3 X 3 kernel in case of 1 channel image and 3 x 3 x n in a n channel image. $\endgroup$ Aug 19, 2022 at 1:18