[GRASS-dev] i.segment at Community Sprint

[Pierre Roudier]

Absolutely Markus. Let's also keep in mind that comparing segmentation
results is a very tricky exercise. There's no such thing as a perfect
segmentation result IMHO.

Pierre

2013/2/12 Markus Metz <markus.metz.giswork at gmail.com>:
> On Mon, Feb 11, 2013 at 8:18 PM, Pierre Roudier
> <pierre.roudier at gmail.com> wrote:
>> Thanks Markus and Eric,
>>
>> I have similar feedback when testing on various SPOT5 scenes. XL results are
>> not entirely identical of course, but reasonably similar, so I support
>> Markus' decision.
>
> I can make the results more similar, but one reason why results are
> different between i.segment and i.segment.xl is that I had the
> impression that i.segment is doing over-segmentation, i.e. merging too
> liberally with the effect that some segments (objects), in particular
> large segments, were more heterogeneous than justified by the given
> merging threshold. With i.segment.xl, I tried to get results that are
> very similar to the eCognition results.
>
> Markus M
>
>>
>> Cheers,
>>
>> Pierre
>>
>> On Feb 11, 2013 10:53 PM, "Markus Metz" <markus.metz.giswork at gmail.com>
>> wrote:
>>>
>>> Hi all,
>>>
>>> I have tested again i.segment and discovered that for the NC Landsat
>>> 2000 imagery the module takes nearly 7 hours for reasonable
>>> segmentation. The i.segment.xl module does the same in 10 seconds. The
>>> region consists of 250 000 cells, not so much. In the documentation it
>>> says that processing the entire ortho image takes about a day. With
>>> the xl version it takes about 10 minutes. I am going to move the xl
>>> version to trunk and keep the original GSoC version in addons for
>>> reference.
>>>
>>> Markus M
>>>
>>>
>>> On Fri, Feb 8, 2013 at 5:47 AM, Eric Momsen <eric.momsen at gmail.com> wrote:
>>> > Sorry for not responding right away, I hadn't realized the code sprint
>>> > started (finished?!) already.  There is a paper deadline this month and
>>> > a
>>> > thesis to write that has consumed my attention...  I don't think I'll
>>> > have
>>> > time for anything in February.  If noone else wants/needs to do it
>>> > sooner, I
>>> > hope to have a break between my thesis writing and starting work this
>>> > summer.  So in April(???) I will find some time to dedicate to GRASS
>>> > code
>>> > and/or documents again.
>>> >
>>> > -Eric
>>> >
>>> >
>>> > On Mon, Feb 4, 2013 at 3:01 AM, Moritz Lennert
>>> > <mlennert at club.worldonline.be> wrote:
>>> >>
>>> >> On 02/02/13 17:37, Markus Metz wrote:
>>> >>>
>>> >>> On Sat, Feb 2, 2013 at 12:56 PM, Moritz Lennert
>>> >>> <mlennert at club.worldonline.be>  wrote:
>>> >>>>
>>> >>>> Markus and Yann,
>>> >>>>
>>> >>>> If you have the time it might be a great opportunity to use the
>>> >>>> community
>>> >>>> sprint to get i.segment from the addons to trunk. What do you think ?
>>> >>>
>>> >>>
>>> >>> I would prefer i.segment.xl for speed reasons and memory control, but
>>> >>> the shape functionality in i.segment would need to be ported to
>>> >>> i.segment.xl first.
>>> >>
>>> >>
>>> >> I meant i.segment in a global sens, not distinguishing .xl, so that's
>>> >> perfectly fine with me.
>>> >>
>>> >> Eric, any chance for you to look at integrating your work on shapes
>>> >> into
>>> >> i.segment.xl ?
>>> >>
>>> >> Moritz
>>> >
>>> >
>>> _______________________________________________
>>> grass-dev mailing list
>>> grass-dev at lists.osgeo.org
>>> http://lists.osgeo.org/mailman/listinfo/grass-dev



--
Scientist
Landcare Research, New Zealand