I was quite confuse about the calculation of the parameters when I learning DL.
So I make one 3D model for VGG16.
each pixel, each activation, each weights equal 1x1x1 size.
Which contain: Volume and Square. ( Volume = Square)
In VGG16 the Memory in Conv3-1 to Conv3-3 all the same size - 56x56x256=802,816
Conv3-2 Volume: 56x56x256
Conv3-2 Square: (56x16)^2
all the data from cs231n lecture9
Most memory is in early CONV
memory vs parameters
Most params are in late FC