[pdal] About litree filter output: "PDAL: All points collinear"

[Ulises Ibarra]

Hi Howard:


El mié, 28 dic 2022 a las 12:45, Howard Butler (<howard at hobu.co>) escribió:

>
>
> > On Dec 27, 2022, at 9:28 PM, Ulises Ibarra <ulisesmartinibarra at gmail.com>
> wrote:
> >
> > In total I have 47 scans of that piece of rainforest. A question: What
> do you think about the processing time of a large file that contains the 47
> scans: Could it take 24 hours X 47 files = 1128 hours?
>
> The AWS spot rate for a c7g.2xlarge in the Oregon region is $0.142. That's
> 8 cpus and 16gb of RAM. Naively splitting your 1128 compute hours over that
> (1128/8 * 0.142) brings up a total cost of $20.002
>
> PDAL purposefully does not split up data and try to internally optimize
> the computing of pipelines because they are extremely sensitive to the
> various filters and their configurations. It is on users to divide and
> conquer on their own with PDAL. For filters.litree, that means breaking the
> data up and trying to find the filters.sample.radius setting that gives you
> good enough results without blowing up memory or computation time.
>

I thought it was my job to send filters.litree a file that already had a
suitable subsampling, something like just a point inside a sphere with
radius 2 cm. I have to read the paper or the litree code to understand, if
I can understand, what's going on there.

>
> Your pipeline with filters.litree is obviously not very efficient, but the
> cost in effort to optimize it for one-time compute jobs far outstrips the
> cost of parallelizing the computation in the cloud somewhere and being done
> with it. That math obviously changes if you need to process the entire
> rainforest with filters.litree :)
>

Thank you Howard.
Happy new year!


>
> Howard
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/pdal/attachments/20230102/969bfe93/attachment.htm>