In lecture Inception layer motivation at 2:00 , Prof Andrew Ng tells that on doing Max-Pool we will have 28 * 28 * 32 output dimension given input was 28 * 28 * 192 dimension.
But in MaxPool no of channels remain same between input and output, so it should be 192 channels in output how it is 32.
Also if we use multiple MaxPool ( let used k no of MaxPool filters ) then on stacking all the outputs from MaxPool we will get 192 * k no of channels in output.
Please tell me what concept I am missing.