TiledViews

Implementation of shifted arrays just like ShiftedArrays.jl but supporting mutation.

Tiled Processing

The function tiled_processing takes an array of input data and dices it up into tiles via constructing a TiledView (see below), and processes each of the tiles (using the eachtile Iterator) with a function provided. The results are then reassembled into a final TiledView whos underlying parent corresponds to an array with the reassembled results. If the tiles are mutually overlapping, appropriate merging via the weights as provided by a window_function are used. This function has to adhere to the interface as defined by the package IndexFunArrays. By default a Hann Window is used.

julia> res = tiled_processing(ones(4,6), (a)->10*a, (2,2), (0,0)).parent
processing tile (1, 1) out of (2, 3), 0.0%
processing tile (2, 1) out of (2, 3), 17.0%
processing tile (1, 2) out of (2, 3), 33.0%
processing tile (2, 2) out of (2, 3), 50.0%
processing tile (1, 3) out of (2, 3), 67.0%
processing tile (2, 3) out of (2, 3), 83.0%
4×6 Matrix{Float64}:
 10.0  10.0  10.0  10.0  10.0  10.0
 10.0  10.0  10.0  10.0  10.0  10.0
 10.0  10.0  10.0  10.0  10.0  10.0
 10.0  10.0  10.0  10.0  10.0  10.0

tiled_processing is especially useful, if the provided function casts the input into a CuArray from the CUDA toolbox and the result is collected back onto the CPU via the Array cast. This allows large arrays, too big for the GPU, to be processed efficiently. For more details on the various ways to use tiled_processing, including functions that return multiple results, see the API of tiled_processing.

TiledViews

A TiledView is a view into the data given upon its construction, which has twice as many dimensions as the original data. The tile size as well as the mutula overlap can be specified. A TiledView works quite seamlessly with a spatial window function.

The number of tiles in a tiled array ta = TiledView(ones(10, 10), (5, 5); keep_center=false) can be accessed by calling get_num_tiles(ta) which in this case results in (2, 2). The tile size (here (5, 5)) as provided upon its construction, can be obtained by the member ta.tile_size.

Here is a small example:

  julia> a = TiledView(reshape(1:49,(7,7)), (4, 4),(1, 1));
  julia> a.parent
  7×7 reshape(::UnitRange{Int64}, 7, 7) with eltype Int64:
   1   8  15  22  29  36  43
   2   9  16  23  30  37  44
   3  10  17  24  31  38  45
   4  11  18  25  32  39  46
   5  12  19  26  33  40  47
   6  13  20  27  34  41  48
   7  14  21  28  35  42  49
   julia> size(a)
  (4, 4, 3, 3)

Note that by default, the center of the central TiledView is forced to agree with the center of the original array. This often leads to more tiles than seemingly necessary. By using the named argument keep_center=false, this forced alignment can be turned off, such that instead the first pixels align. For more details on the various ways to construct, see the API documentation.