[pdal] Way to cast dimension data type?

[Piero Toffanin]

Hello,

First let me say that PDAL rocks. I love it.

I was wondering if there's a way to specify (cast) the data type of the 
X,Y,Z dimensions while reading a PLY file (so in effect override the 
type specified in the PLY header) in the PDAL pipeline.

I have an input PLY file (untransformed.ply) that has the header:

ply
format ascii 1.0
[ .... ]
property float x
property float y
property float z
[ .... ]

And coordinate points with ~5 digits of precision.

1.13759 12.1322 162.786
[ .... ]

When I try to apply the following transform (add a east/north offset):

{
   "pipeline": [
     "untransformed.ply",
     {
       "matrix": "1 0 0 576705 0 1 0 5188170 0 0 1 0 0 0 0 1",
       "type": "filters.transformation"
     },
     {
       "a_srs": "EPSG:32615",
       "filename": "transformed.las",
       "offset_x": "576705",
       "offset_y": "5188170",
       "offset_z": "0"
     }
   ]
}

There's a loss of precision (I get a "grid-like" output, see 
https://cloud.githubusercontent.com/assets/1951843/25507885/9cd01d12-2b7c-11e7-9c54-0e219dae4f05.png). 
I think it's because the matrix transformation code here 
https://github.com/PDAL/PDAL/blob/dd97a033c5d487000e200304a7619e9978de6ae2/filters/TransformationFilter.cpp#L106 
_correctly_ treats points as floats.

Changing the PLY header to:

ply
format ascii 1.0
[ .... ]
property double x
property double y
property double z
[ .... ]

Produces the expected output (no grid-like output).

I was wondering if I could force the dimensions to be treated as a 
double directly from PDAL? Or perhaps is there interest in adding a flag 
(cast to dimension) in PDAL's PLY reader (I'd be happy to contribute to it)?

More information about the background of this issue is also available 
here: https://github.com/OpenDroneMap/OpenDroneMap/issues/522

Thank you for any information you might have!

-Piero

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/pdal/attachments/20170506/878d9856/attachment.html>