Hi,
Working with a colleague and comparing results we were wondering whether i.segment contains any randomness effects which might make results differ from one run to another.
In the manual it says in section Seeds in the chapter on region growing:
"The seeds map can be used to provide either seed pixels (random or selected points from which to start the segmentation process) or seed segments."
which sounds as if the module creates random seeds if none are given. Looking through the code, I do not find such creation of random seeds, though. Is this just me not looking correctly, or aren't there any ?
Moritz
On Wed, Mar 28, 2018 at 3:04 PM, Moritz Lennert <mlennert@club.worldonline.be> wrote:
Hi,
Working with a colleague and comparing results we were wondering whether i.segment contains any randomness effects which might make results differ from one run to another.
There is no randomness in i.segments. Given he same input data and settings, results should be identical between runs, i.e. repeatable.
Did you do a simple diff with r.mapcalc?
In the manual it says in section Seeds in the chapter on region growing:
“The seeds map can be used to provide either seed pixels (random or selected points from which to start the segmentation process) or seed segments.”
which sounds as if the module creates random seeds if none are given. Looking through the code, I do not find such creation of random seeds, though. Is this just me not looking correctly, or aren’t there any ?
A seed map can be created by randomly selecting pixels, e.g. with r.random. If a seed map is not given, all non-NULL cells are seeds.
HTH,
Markus M
Moritz
grass-dev mailing list
grass-dev@lists.osgeo.org
https://lists.osgeo.org/mailman/listinfo/grass-dev
On 28/03/18 17:53, Markus Metz wrote:
On Wed, Mar 28, 2018 at 3:04 PM, Moritz Lennert <mlennert@club.worldonline.be <mailto:mlennert@club.worldonline.be>> wrote:
>
> Hi,
>
> Working with a colleague and comparing results we were wondering whether i.segment contains any randomness effects which might make results differ from one run to another.
There is no randomness in i.segments. Given he same input data and settings, results should be identical between runs, i.e. repeatable.
Did you do a simple diff with r.mapcalc?
No, but we have variance and spatial autocorrelation values for the datasets and didn't get the same results. As there is no randomness, I suppose my colleague did not define the region the same way I did. Unfortunately, he only kept the numerical values, not the segmentation results.
>
> In the manual it says in section Seeds in the chapter on region growing:
>
> "The seeds map can be used to provide either seed pixels (random or selected points from which to start the segmentation process) or seed segments."
>
> which sounds as if the module creates random seeds if none are given. Looking through the code, I do not find such creation of random seeds, though. Is this just me not looking correctly, or aren't there any ?
A seed map can be created by randomly selecting pixels, e.g. with r.random. If a seed map is not given, all non-NULL cells are seeds.
Thanks for the info.
Moritz