# writers.bpf¶

BPF is an NGA specification for point cloud data. The specification can be found at https://nsgreg.nga.mil/doc/view?i=4202 The PDAL BPF Writer only supports writing of version 3 BPF format files.

## Example¶

{
"pipeline":[
{
"filename":"inputfile.las"
},
{
"type":"writers.bpf",
"filename":"outputfile.bpf"
}
]
}


## Options¶

filename
BPF file to read. The writer will accept a filename containing a single placeholder character (‘#’). If input to the writer consists of multiple PointViews, each will be written to a separate file, where the placeholder will be replaced with an incrementing integer. If no placeholder is found, all PointViews provided to the writer are aggregated into a single file for output. Multiple PointViews are usually the result of using filters.splitter, filters.chipper or filters.divider. [Required]
compression
This option can be set to true to cause the file to be written with Zlib compression as described in the BPF specification. [Default: false]
format

Specifies the format for storing points in the file. [Default: dim]

• dim == Dimension-major (non-interleaved). All data for a single dimension are stored contiguously.
• point == Point-major (interleaved). All data for a single point are stored contiguously.
• byte == Byte-major (byte-segregated). All data for a single dimension are stored contiguously, but bytes are arranged such that the first bytes for all points are stored contiguously, followed by the second bytes of all points, etc. See the BPF specification for further information.
bundledfile
Path of file to be written as a bundled file (see specification). The path part of the filespec is removed and the filename is stored as part of the data. This option can be specified as many times as desired.
Base64-encoded data that will be decoded and written following the standard BPF header.
coord_id
The coordinate ID (UTM zone) of the data. NOTE: Only the UTM coordinate type is currently supported. [Default: 0, with coordinate type set to none]
scale_x, scale_y, scale_z

Scale to be divided from the X, Y and Z nominal values, respectively, after the offset has been applied. The special value “auto” can be specified, which causes the writer to select a scale to set the stored values of the dimensions to range from [0, 2147483647]. [Default: .01]

Note

written value = (nominal value - offset) / scale.

offset_x, offset_y, offset_z

Offset to be subtracted from the X, Y and Z nominal values, respectively, before the value is scaled. The special value “auto” can be specified, which causes the writer to set the offset to the minimum value of the dimension. [Default: auto]

Note

written value = (nominal value - offset) / scale.

Note

Because BPF data is always stored in UTM, the XYZ offsets are set to “auto” by default. This is to avoid truncation of the decimal digits (which may occur with offsets left at 0).

output_dims
If specified, limits the dimensions written for each point. Dimensions are listed by name and separated by commas. X, Y and Z are required and must be explicitly listed.