[GRASS-dev] i.segment error: Ri is 0

[Markus Metz]

On Tue, Jul 2, 2013 at 9:58 AM, Moritz Lennert
<mlennert at club.worldonline.be> wrote:
> On 01/07/13 21:04, Markus Metz wrote:
>>
>> On Mon, Jul 1, 2013 at 3:56 PM, Moritz Lennert
>> <mlennert at club.worldonline.be>  wrote:
>>>
>>> Hi Eric and Markus,
>>>
>>> Trying to use i.segment in grass7 checked out and compiled a few days ago
>>> (rev 56918), I came upon the following error and the resulting segments
>>> file
>>> was not created. I can file this as a bug report, but wanted your
>>> feedback
>>> first to see if I'm misusing i.segment somehow. I haven't been able to
>>> find
>>> the error in the source code.
>>>
>>> command line:
>>> time i.segment group=xs out=seg_xs minsize=2 memory=3072 threshold=0.2
>>> --o
>>>
>>> error:
>>> Segmentation converged after 16 iterations.
>>> Merging segments smaller than 2 cells
>>> ERREUR :Ri is 0
>>
>>
>> This should not happen. The ID of a segment is always positive or
>> negative or NULL (Rast_is_c_null_value())
>>>
>>>
>>> This is on a mosaic of Worldview 2 images with region specs as follows:
>>>
>>>> g.region -p
>>>
>>> projection: 1 (UTM)
>>> zone:       33
>>> datum:      wgs84
>>> ellipsoid:  wgs84
>>> north:      4876400
>>> south:      4849792
>>> west:       610056
>>> east:       634648
>>> nsres:      2
>>> ewres:      2
>>> rows:       13304
>>> cols:       12296
>>> cells:      163585984
>>>
>>> The mosaic is only a narrow band within that region, so that actually
>>> there
>>> are only 34,755,878 non-null cells.
>>>
>>> Any hints ?
>>
>>
>> Not really. I created a sample dataset with a MASK leaving only a
>> narrow diagonal strip and everything went fine.
>
>
> I did not use a mask, the non-relevant cells were just nulls. Would it be
> better to create a mask ?

No, i.segment tests with Rast_is_d_null_value() which is true for both
NULL cells and MASK'ed cells.
>>
>> Can you provide data to replicate or commands using one of the sample
>> datasets to replicate this error?
>
>
> I'll run the segmentation again without having something else running. If I
> can reproduce the error, I'll make the data available to you so you can
> test.

Other processes should not be able to interfere, as long as enough
system resources are available which seems to be the case. I suspect a
problem with nodata handling, but the code looks ok to me.

If need be, you can send me (a link to) the test data off-list.

Markus M