Update conv.jl

Specifying the format of the data given to the convolutional layer.

Author: vivekkumar7089

src/layers/conv.jl

@@ -40,7 +40,8 @@ Image data should be stored in WHCN order (width, height, channels, batch).
 In other words, a 100×100 RGB image would be a `100×100×3×1` array,
 and a batch of 50 would be a `100×100×3×50` array.
 This has `N=2` spatial dimensions, and needs a kernel size like `(5,5)`,
-a 2-tuple of integers.
+a 2-tuple of integers. For different `N`, data should be `N_1 x N_2 x ... x N_n x channels x batch`,
+i.e. `spatial/temporal dimensions x channels x batch`.
 
 For `N` spatial dimensions, this layer expects as input an array
 with `ndims(x) == N+2`, where `size(x,N+1) == in` is the number of channels.