Flux.jl, treat rows independently

I would like to train on the rows feature independently. Therefore, I would like to have a dense layer for row 1, row 2 …, row n and at the end combine the 12 row outputs. Is it possible in Flux.jl. At the end I would like to do the same with a CNN. Thx for any clue.

´´´
x=rand(1000,12,2000)
Chain(Dense(…))
´´´