(using the imagery group created in i.cluster example, otherwise unrelated)
I looked at goodness from i.segment and then reject from i.maxlik. I noticed that there is some correlation between these two. Well, this is quite okay since they are using same pixel values. However, then I noticed that low values of goodness (-5000, …) correlate with low values of reject (1 == 1%). But the i.maxlik manual says that “1 = keep and 16 = reject”. Can somebody explain this to me and perhaps suggest improvement to the manual? Shouldn’t high rejection correlate with low goodness?
And also, less important thing, i.maxlik reject is an categorical map with percentages reclassified to 16 classes. Does this have some practical meaning or is this just an over-left from integer-only times?
2014-11-16 3:21 GMT+01:00 Vaclav Petras <wenzeslaus@gmail.com>:
And also, less important thing, i.maxlik reject is an categorical map with
percentages reclassified to 16 classes. Does this have some practical
meaning or is this just an over-left from integer-only times?
I would guess that it's over-left. It would be nice to synchronize
this output (reject vs. goodness) for i.maxlik and i.smap... Ideally
before releasing GRASS 7...
(using the imagery group created in i.cluster example, otherwise unrelated)
I looked at goodness from i.segment and then reject from i.maxlik. I noticed
that there is some correlation between these two. Well, this is quite okay
since they are using same pixel values. However, then I noticed that low
values of goodness (-5000, ...) correlate with low values of reject (1 ==
1%). But the i.maxlik manual says that "1 = keep and 16 = reject". Can
somebody explain this to me and perhaps suggest improvement to the manual?
Shouldn't high rejection correlate with low goodness?
Unless someone looks at the source code, the i.maxlik manual may
simply be wrong.
A nice, synthetic test case would be the best.
And also, less important thing, i.maxlik reject is an categorical map with
percentages reclassified to 16 classes. Does this have some practical
meaning or is this just an over-left from integer-only times?
(using the imagery group created in i.cluster example, otherwise unrelated)
I looked at goodness from i.segment and then reject from i.maxlik. I noticed
that there is some correlation between these two. Well, this is quite okay
since they are using same pixel values. However, then I noticed that low
values of goodness (-5000, ...)
i.segment's goodness estimate is supposed to be in the range [0,1].
Fixed in r62793,4. Can you test again?
Markus M
correlate with low values of reject (1 ==
1%). But the i.maxlik manual says that "1 = keep and 16 = reject". Can
somebody explain this to me and perhaps suggest improvement to the manual?
Shouldn't high rejection correlate with low goodness?
And also, less important thing, i.maxlik reject is an categorical map with
percentages reclassified to 16 classes. Does this have some practical
meaning or is this just an over-left from integer-only times?
(using the imagery group created in i.cluster example, otherwise unrelated)
I looked at goodness from i.segment and then reject from i.maxlik. I noticed
that there is some correlation between these two. Well, this is quite okay
since they are using same pixel values. However, then I noticed that low
values of goodness (-5000, ...)
i.segment's goodness estimate is supposed to be in the range [0,1].
Fixed in r62793,4. Can you test again?
Could you explain / put into the manual the explanation of the calculation / meaning of this goodness of fit ?
(using the imagery group created in i.cluster example, otherwise
unrelated)
I looked at goodness from i.segment and then reject from i.maxlik. I
noticed
that there is some correlation between these two. Well, this is quite
okay
since they are using same pixel values. However, then I noticed that low
values of goodness (-5000, ...)
i.segment's goodness estimate is supposed to be in the range [0,1].
Fixed in r62793,4. Can you test again?
Could you explain / put into the manual the explanation of the calculation /
meaning of this goodness of fit ?
Done in r62830:
The goodness of fit for each pixel is calculated as 1 - distance
of the pixel to the object it belongs to. The distance is calculated
with the selected similarity method. A value of 1 means
identical values, perfect fit, and a value of 0 means maximum possible
distance, worst possible fit.
(using the imagery group created in i.cluster example, otherwise
unrelated)
I looked at goodness from i.segment and then reject from i.maxlik. I
noticed
that there is some correlation between these two. Well, this is quite
okay
since they are using same pixel values. However, then I noticed that low
values of goodness (-5000, ...)
i.segment's goodness estimate is supposed to be in the range [0,1].
Fixed in r62793,4. Can you test again?
Could you explain / put into the manual the explanation of the calculation /
meaning of this goodness of fit ?
Done in r62830:
The goodness of fit for each pixel is calculated as 1 - distance
of the pixel to the object it belongs to. The distance is calculated
with the selected similarity method. A value of 1 means
identical values, perfect fit, and a value of 0 means maximum possible
distance, worst possible fit.
