i.segment.uspo produces empty output

[mwicht]

Hi @tgrippa,
I am trying to use your code for a project of mine. I am working with pleiades imagery for optical and LiDAR for nDSM.
So far the first part runs smoothly but I started encountering problems when running i.segment.uspo. It produces an empty *.csv file, no matter the parameters (thresholds/minsize).
I can see that all datasets and subsets are created within the GRASS Mapset TEST_USPO.

I have a shapefile with one polygon per landcover class which is used as regions in i.segment.uspo.

The regions are printed correctly as well:
subset_uspo_1@TEST_USPO,subset_uspo_10@TEST_USPO,subset_uspo_11@TEST_USPO,subset_uspo_12@TEST_USPO,subset_uspo_2@TEST_USPO,subset_uspo_3@TEST_USPO,subset_uspo_4@TEST_USPO,subset_uspo_5@TEST_USPO,subset_uspo_6@TEST_USPO,subset_uspo_7@TEST_USPO,subset_uspo_8@TEST_USPO,subset_uspo_9@TEST_USPO

Am I assuming right, that the single polygons are supposed to define one class each?
Or could something else be an issue here?

Did you encounter issues like this during the development? Any hints on debugging this issue?

[tgrippa]

Hi @mwicht,
Did you try to run i.segment.uspo directly in the GRASS GUI, with the same parameters as in your jupyter notebook ? Is it also producing a empty .csv ? If yes, are you sure the range of thresholds you ask i.segment.uspo to look into is adapted to you data ? Be sure that the 'threshold_start' and 'threshold_stop' well correspond to situations of under and over-segmentation, respectively.
Also, I see that you have some regions that are very small. Maybe the problem come from that and you should consider to increase a bit their area.

[mwicht]

Hi @tgrippa,
yes, I tried running i.segment.uspo in the GRASS GUI and still obtaining empty outputs. The same happens with i.segment.
I am running i.segment.uspo with a wide range of parameter thresholds. As I understand, this function will obtain the best settings for segmentation (i.segment) by iterating through the thresholds using the threshold step. So if I start at a threshold of e.g. 0.05 and stop at 0.99 with a step of 0.05, I should get results at some point, right?
If I just run i.segment it will tell me at the end at the command output how many segments are created. But the obtained raster map is always empty (visually, histogram).
How can I adapt the parameters to my data? What would be the least restrictive settings which HAVE to produce some output which I can base my refinement on? The mentioned thresholds in the i.segment documentation of 0.01 to 0.05 did not work either.

Attached you can find a small subset of my data which will not produce any output. Could you please have a look at this, if you can spare the time? Thank you!

test_area.zip

[tgrippa]

I just tried and it works on my computer. The best threshold according to i.segment.uspo is 0.10. I ask to test the following range : 0.02 until 0.30 with a step of 0.02. I fixed the minsize to 15. The result looks great.

I'm wondering if your problem doesn't come to the fact that your regions in which you ask i.segment.uspo to test are badly set at some point of the processing.
What is the result of the following command ? g.region -p region=subset_uspo_1 ?

[tgrippa]

Hi @mwicht,
Did you solve your issue since your last comment ? Can we consider this issue close ?

[Doctor-Who]

Hi everyone,
Same error : with GRASS GUI it works on a Linux Laptop but when trying launch into a docker instance on a computing instance, empty output (volume, etc. everything is ok as i.segment is working)

[tgrippa]

Hi @Doctor-Who ! I'm not familiar at all with Docker so I could not help you :/

[Doctor-Who]

Thanks for your reply, I will try in Jupiter Notebook