[GRASS-dev] Testing i.superpixels.slic

Moritz Lennert mlennert at club.worldonline.be
Thu Mar 2 08:52:22 PST 2017


On 02/03/17 12:09, Sophie Crommelinck wrote:
> 2017-03-01 16:45 GMT+01:00 Moritz Lennert <mlennert at club.worldonline.be
> <mailto:mlennert at club.worldonline.be>>:
>
> I also tested increasing the number of iterations to 100. This resulted
> in a lower number of tiny superpixels, but still some superpixels are
> far away from being equally shaped compared to the rest.

If you want more equal shape, you have to increase compactness. And you 
can try the minsize option. In the original version, they implement a 
minsize internally, but then merge superpixels that are smaller than 
that minsize to an arbitrarily chosen neighbor. In i.superpixels.slic, 
the merging is done to the most similar neighbor.

> Also this
> resulted in a processing time of several hours.

Make sure to set the memory parameter to as much memory as you can 
afford. But in tests between MarkusM and me we also saw that the speed 
of the CPU makes a significant difference.

>     In short: don't expect similar results using same settings. Just
>     experiment with i.superpixels.slic until you get a satisfying result.
>
>     Also don't forget that in GRASS GIS you can input an
>
>
> Where did the rest of the sentence go?

Somewhere in the void ;-)

I wanted to point out that contrary to the original version, which is 
limited to three RGB layers, the GRASS version takes any arbitrary 
number of input layers. So, if your data is RGB + NIR for example, you 
can input all 4, and you can also add texture layers, NDVI, or others.

This can potentially improve your results. But it will also increase 
computing time.

Moritz


More information about the grass-dev mailing list