Effect of stride length on depth of the output of a convolution

In Lesson 6, Jeremy said increasing the stride increases the depth of the output.
But isn’t the depth based on the number of kernels/filters you are using?

Would you please provide the time point and exact quote?

Sure. Around 1:25

He’s saying that when you increase the stride, you can choose to increase the number of filters (depth) while using the same amount of memory for the activations. The spatial dimension is reduced, which allows the filters/kernels/features/depth to be increased. The increase of depth does not happen automatically.

Good question that will also help others to understand.


Thanks. That clears it up!