I have a working solution for ensuring that the created point is placed within the polygon when using --add-pois-to-areas, based on drawing the polygon on to a small monochrome bitmap and then looking for the point that is furthest from the surrounding area. I used a 9x9 bitmap for polygons having a small number of points and 15x15 for longer polygons. There is however a performance penalty. My standard map takes about 1 hour 20 minutes; using this algorithm the time increases by about 50% to about 2 hours. I am not currently able to commit changes to SVN (perhaps someone can help out with that) but I have attached the code changes. I suggest that due to the performance penalty, if we adopt this, then the --add-pois-to-areas option be extended to be --add-pois-to-areas[=centre|optimised] or something similar, with the default being centre and functioning as now and the optimised option invoking the new code. Please try out the suggested change. Note I don't expect this to work properly where a polygon is formed from a multiploygon relation, but the code could quite easily be adapted for this circumstance. Regards, Mike
Hi Mike, 50% sounds better than my algo, but still quite a lot. I'll have a closer look at your algo later. Please note that your change has a side effect on the house number generator. Up to now this doesn't contain a filter for generated POI, so each polygon with a house number is processed twice, once because of the POI, once because the Generator uses Way.getCofG(). If both have different positions this might have a negative impact. Gerd From: mike@tvage.co.uk To: mkgmap-dev@lists.mkgmap.org.uk Date: Tue, 6 Jan 2015 14:56:30 +0000 Subject: Re: [mkgmap-dev] small issue with Way.getCofG() I have a working solution for ensuring that the created point is placed within the polygon when using --add-pois-to-areas, based on drawing the polygon on to a small monochrome bitmap and then looking for the point that is furthest from the surrounding area. I used a 9x9 bitmap for polygons having a small number of points and 15x15 for longer polygons. There is however a performance penalty. My standard map takes about 1 hour 20 minutes; using this algorithm the time increases by about 50% to about 2 hours. I am not currently able to commit changes to SVN (perhaps someone can help out with that) but I have attached the code changes. I suggest that due to the performance penalty, if we adopt this, then the --add-pois-to-areas option be extended to be --add-pois-to-areas[=centre|optimised] or something similar, with the default being centre and functioning as now and the optimised option invoking the new code. Please try out the suggested change. Note I don't expect this to work properly where a polygon is formed from a multiploygon relation, but the code could quite easily be adapted for this circumstance. Regards, Mike _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
Hi Mike, I like the idea, but it seems to be slow. Is it possible that your algo suffers when no fast graphics card is available? On my netbook the performance is very poor, did not yet try on the PC, but that also has no high speed graphics. Gerd GerdP wrote
Hi Mike,
50% sounds better than my algo, but still quite a lot. I'll have a closer look at your algo later. Please note that your change has a side effect on the house number generator. Up to now this doesn't contain a filter for generated POI, so each polygon with a house number is processed twice, once because of the POI, once because the Generator uses Way.getCofG(). If both have different positions this might have a negative impact.
Gerd
From:
mike@.co
To:
mkgmap-dev@.org
Date: Tue, 6 Jan 2015 14:56:30 +0000 Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
I have a working solution for ensuring that the created point is placed within the polygon when using --add-pois-to-areas, based on drawing the polygon on to a small monochrome bitmap and then looking for the point that is furthest from the surrounding area. I used a 9x9 bitmap for polygons having a small number of points and 15x15 for longer polygons. There is however a performance penalty. My standard map takes about 1 hour 20 minutes; using this algorithm the time increases by about 50% to about 2 hours. I am not currently able to commit changes to SVN (perhaps someone can help out with that) but I have attached the code changes. I suggest that due to the performance penalty, if we adopt this, then the --add-pois-to-areas option be extended to be --add-pois-to-areas[=centre|optimised] or something similar, with the default being centre and functioning as now and the optimised option invoking the new code. Please try out the suggested change. Note I don't expect this to work properly where a polygon is formed from a multiploygon relation, but the code could quite easily be adapted for this circumstance.
Regards, Mike
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev _______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
-- View this message in context: http://gis.19327.n5.nabble.com/small-issue-with-Way-getCofG-tp5828821p582924... Sent from the Mkgmap Development mailing list archive at Nabble.com.
Hi Mike, on my PC the performance of your algo is quite good. Attached is a patch that contains your patch as well as my quick implementation of the algo described here: http://arxiv.org/ftp/arxiv/papers/1212/1212.3193.pdf The patch tests only performance, it computes the center with the 3 different algos, I've commented the part that prints times and GPX data for debug purposes. I noticed that the results between both algos are very different, I did not yet try to find out which one is better, but mine is much slower on my PC. I also noticed that your algo doesn't always calculate a point in the polygon, see e.g. way 178708143. If you like, please try to find a better compromise, I like to fix a problem in splitter first. I also did not yet look at the effect on the house number code, as there are many more small open problems, but I think it should be easy to sort that out. Gerd
Date: Tue, 6 Jan 2015 13:23:57 -0700 From: gpetermann_muenchen@hotmail.com To: mkgmap-dev@lists.mkgmap.org.uk Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
Hi Mike,
I like the idea, but it seems to be slow. Is it possible that your algo suffers when no fast graphics card is available? On my netbook the performance is very poor, did not yet try on the PC, but that also has no high speed graphics.
Gerd
GerdP wrote
Hi Mike,
50% sounds better than my algo, but still quite a lot. I'll have a closer look at your algo later. Please note that your change has a side effect on the house number generator. Up to now this doesn't contain a filter for generated POI, so each polygon with a house number is processed twice, once because of the POI, once because the Generator uses Way.getCofG(). If both have different positions this might have a negative impact.
Gerd
From:
mike@.co
To:
mkgmap-dev@.org
Date: Tue, 6 Jan 2015 14:56:30 +0000 Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
I have a working solution for ensuring that the created point is placed within the polygon when using --add-pois-to-areas, based on drawing the polygon on to a small monochrome bitmap and then looking for the point that is furthest from the surrounding area. I used a 9x9 bitmap for polygons having a small number of points and 15x15 for longer polygons. There is however a performance penalty. My standard map takes about 1 hour 20 minutes; using this algorithm the time increases by about 50% to about 2 hours. I am not currently able to commit changes to SVN (perhaps someone can help out with that) but I have attached the code changes. I suggest that due to the performance penalty, if we adopt this, then the --add-pois-to-areas option be extended to be --add-pois-to-areas[=centre|optimised] or something similar, with the default being centre and functioning as now and the optimised option invoking the new code. Please try out the suggested change. Note I don't expect this to work properly where a polygon is formed from a multiploygon relation, but the code could quite easily be adapted for this circumstance.
Regards, Mike
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev _______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
-- View this message in context: http://gis.19327.n5.nabble.com/small-issue-with-Way-getCofG-tp5828821p582924... Sent from the Mkgmap Development mailing list archive at Nabble.com. _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
HI Gerd, the method I've used won't work if the polygon is thinner than the resolution of the bitmap. I hadn't considered that it might be used on rivers; my main testing has been on car parks, which although they may be irregularly shaped are not normally very long and very thin (I handle river names from the underlying lines). There are two possible problems that could occur - if a polygon is thin enough that it renders in the bitmap as a line that is nowhere wider than 2 pixels the POI could be placed just outside the polygon. I haven't checked what fillPolygon does if the points have no volume and form a line. If this renders nothing, then the algo will just return the centre point, which could be nowhere near the polygon. A simple solution to the first problem would be to examine the biggestSquaredDistanceToBlack figure before returning the coordinate and if the value is 1, repeat the operation on a small area around the point to get a better accuracy. I'll look into that. For the second problem I'll add an error message, so we can determine if it actually occurs (biggestSquaredDistanceToBlack is 0). I've also thought that having drawn the bitmap, it might be more efficient to then copy its contents out into an array so that finding the distances from pixels inside the polygon to those outside become simple array lookups instead of graphics function calls. I'll look into that as well. Cheers, Mike From: Gerd Petermann [mailto:gpetermann_muenchen@hotmail.com] Sent: 07 January 2015 06:55 To: mkgmap-dev@lists.mkgmap.org.uk Subject: Re: [mkgmap-dev] small issue with Way.getCofG() Hi Mike, on my PC the performance of your algo is quite good. Attached is a patch that contains your patch as well as my quick implementation of the algo described here: http://arxiv.org/ftp/arxiv/papers/1212/1212.3193.pdf The patch tests only performance, it computes the center with the 3 different algos, I've commented the part that prints times and GPX data for debug purposes. I noticed that the results between both algos are very different, I did not yet try to find out which one is better, but mine is much slower on my PC. I also noticed that your algo doesn't always calculate a point in the polygon, see e.g. way 178708143. If you like, please try to find a better compromise, I like to fix a problem in splitter first. I also did not yet look at the effect on the house number code, as there are many more small open problems, but I think it should be easy to sort that out. Gerd
Date: Tue, 6 Jan 2015 13:23:57 -0700 From: gpetermann_muenchen@hotmail.com <mailto:gpetermann_muenchen@hotmail.com> To: mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk>
Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
Hi Mike,
I like the idea, but it seems to be slow. Is it possible that your algo suffers when no fast graphics card is available? On my netbook the performance is very poor, did not yet try on the PC, but that also has no high speed graphics.
Gerd
GerdP wrote
Hi Mike,
50% sounds better than my algo, but still quite a lot. I'll have a closer look at your algo later. Please note that your change has a side effect on the house number generator. Up to now this doesn't contain a filter for generated POI, so each polygon with a house number is processed twice, once because of the POI, once because the Generator uses Way.getCofG(). If both have different positions this might have a negative impact.
Gerd
From:
mike@.co <mailto:mike@.co>
To:
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
Date: Tue, 6 Jan 2015 14:56:30 +0000 Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
I have a working solution for ensuring that the created point is placed within the polygon when using --add-pois-to-areas, based on drawing the polygon on to a small monochrome bitmap and then looking for the point that is furthest from the surrounding area. I used a 9x9 bitmap for polygons having a small number of points and 15x15 for longer polygons. There is however a performance penalty. My standard map takes about 1 hour 20 minutes; using this algorithm the time increases by about 50% to about 2 hours. I am not currently able to commit changes to SVN (perhaps someone can help out with that) but I have attached the code changes. I suggest that due to the performance penalty, if we adopt this, then the --add-pois-to-areas option be extended to be --add-pois-to-areas[=centre|optimised] or something similar, with the default being centre and functioning as now and the optimised option invoking the new code. Please try out the suggested change. Note I don't expect this to work properly where a polygon is formed from a multiploygon relation, but the code could quite easily be adapted for this circumstance.
Regards, Mike
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev _______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
-- View this message in context: http://gis.19327.n5.nabble.com/small-issue-with-Way-getCofG-tp5828821p582924 7.html Sent from the Mkgmap Development mailing list archive at Nabble.com. _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
Hi Mike, reg. POI for polygons like rivers etc: The current algo calculates a POI for each polygon before any style rule is applied. If the calculation of the Coord takes a lot of time, and many of the POI are later dropped in the points rules, maybe it is possible to create a dummy object first, pass it through the style rules, check if it produces a map object, and calculate the exact position only for those. I think we would have to change/remove the POIGeneratorHook and call it "on the fly", but I don't see any big problem with that. Gerd From: mike@tvage.co.uk To: mkgmap-dev@lists.mkgmap.org.uk Date: Wed, 7 Jan 2015 09:29:51 +0000 Subject: Re: [mkgmap-dev] small issue with Way.getCofG() HI Gerd, the method I’ve used won’t work if the polygon is thinner than the resolution of the bitmap. I hadn’t considered that it might be used on rivers; my main testing has been on car parks, which although they may be irregularly shaped are not normally very long and very thin (I handle river names from the underlying lines). There are two possible problems that could occur – if a polygon is thin enough that it renders in the bitmap as a line that is nowhere wider than 2 pixels the POI could be placed just outside the polygon. I haven’t checked what fillPolygon does if the points have no volume and form a line. If this renders nothing, then the algo will just return the centre point, which could be nowhere near the polygon. A simple solution to the first problem would be to examine the biggestSquaredDistanceToBlack figure before returning the coordinate and if the value is 1, repeat the operation on a small area around the point to get a better accuracy. I’ll look into that. For the second problem I’ll add an error message, so we can determine if it actually occurs (biggestSquaredDistanceToBlack is 0). I’ve also thought that having drawn the bitmap, it might be more efficient to then copy its contents out into an array so that finding the distances from pixels inside the polygon to those outside become simple array lookups instead of graphics function calls. I’ll look into that as well. Cheers,Mike From: Gerd Petermann [mailto:gpetermann_muenchen@hotmail.com] Sent: 07 January 2015 06:55 To: mkgmap-dev@lists.mkgmap.org.uk Subject: Re: [mkgmap-dev] small issue with Way.getCofG() Hi Mike, on my PC the performance of your algo is quite good. Attached is a patch that contains your patch as well as my quick implementation of the algo described here: http://arxiv.org/ftp/arxiv/papers/1212/1212.3193.pdf The patch tests only performance, it computes the center with the 3 different algos, I've commented the part that prints times and GPX data for debug purposes. I noticed that the results between both algos are very different, I did not yet try to find out which one is better, but mine is much slower on my PC. I also noticed that your algo doesn't always calculate a point in the polygon, see e.g. way 178708143. If you like, please try to find a better compromise, I like to fix a problem in splitter first. I also did not yet look at the effect on the house number code, as there are many more small open problems, but I think it should be easy to sort that out. Gerd> Date: Tue, 6 Jan 2015 13:23:57 -0700
From: gpetermann_muenchen@hotmail.com To: mkgmap-dev@lists.mkgmap.org.uk Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
Hi Mike,
I like the idea, but it seems to be slow. Is it possible that your algo suffers when no fast graphics card is available? On my netbook the performance is very poor, did not yet try on the PC, but that also has no high speed graphics.
Gerd
GerdP wrote
Hi Mike,
50% sounds better than my algo, but still quite a lot. I'll have a closer look at your algo later. Please note that your change has a side effect on the house number generator. Up to now this doesn't contain a filter for generated POI, so each polygon with a house number is processed twice, once because of the POI, once because the Generator uses Way.getCofG(). If both have different positions this might have a negative impact.
Gerd
From:
mike@.co
To:
mkgmap-dev@.org
Date: Tue, 6 Jan 2015 14:56:30 +0000 Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
I have a working solution for ensuring that the created point is placed within the polygon when using --add-pois-to-areas, based on drawing the polygon on to a small monochrome bitmap and then looking for the point that is furthest from the surrounding area. I used a 9x9 bitmap for polygons having a small number of points and 15x15 for longer polygons. There is however a performance penalty. My standard map takes about 1 hour 20 minutes; using this algorithm the time increases by about 50% to about 2 hours. I am not currently able to commit changes to SVN (perhaps someone can help out with that) but I have attached the code changes. I suggest that due to the performance penalty, if we adopt this, then the --add-pois-to-areas option be extended to be --add-pois-to-areas[=centre|optimised] or something similar, with the default being centre and functioning as now and the optimised option invoking the new code. Please try out the suggested change. Note I don't expect this to work properly where a polygon is formed from a multiploygon relation, but the code could quite easily be adapted for this circumstance.
Regards, Mike
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev _______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org
-- View this message in context: http://gis.19327.n5.nabble.com/small-issue-with-Way-getCofG-tp5828821p582924... Sent from the Mkgmap Development mailing list archive at Nabble.com. _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
Hi Gerd, I think there are many drawbacks to move the POI generation into the style code. At least it will complicate the style code. Maybe it is better to implement your idea directly in the POIGeneratorHook. If it has access to the point style it could check itself if the polygon will (probably) create a POI. Maybe it could first test if there is any overlap of the polygon tags with the tags used in the points file. Then it could pass an artifical POI through the points style and check if a POI is created (finally rules might be disabled to speed up the processing). The hook might also decide how to proceed dependent on the complexity of the polygon. By the way: Did you think of having different methods depending on the use case? So one getCofG() for POI placement, one getHouseNumberPoint() for housenumber processing etc. It sounds like it is hard to find one algorithm that fits all. WanMil
Hi Mike,
reg. POI for polygons like rivers etc:
The current algo calculates a POI for each polygon before any style rule is applied. If the calculation of the Coord takes a lot of time, and many of the POI are later dropped in the points rules, maybe it is possible to create a dummy object first, pass it through the style rules, check if it produces a map object, and calculate the exact position only for those. I think we would have to change/remove the POIGeneratorHook and call it "on the fly", but I don't see any big problem with that.
Gerd
------------------------------------------------------------------------ From: mike@tvage.co.uk To: mkgmap-dev@lists.mkgmap.org.uk Date: Wed, 7 Jan 2015 09:29:51 +0000 Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
HI Gerd, the method I’ve used won’t work if the polygon is thinner than the resolution of the bitmap. I hadn’t considered that it might be used on rivers; my main testing has been on car parks, which although they may be irregularly shaped are not normally very long and very thin (I handle river names from the underlying lines). There are two possible problems that could occur – if a polygon is thin enough that it renders in the bitmap as a line that is nowhere wider than 2 pixels the POI could be placed just outside the polygon. I haven’t checked what fillPolygon does if the points have no volume and form a line. If this renders nothing, then the algo will just return the centre point, which could be nowhere near the polygon. A simple solution to the first problem would be to examine the biggestSquaredDistanceToBlack figure before returning the coordinate and if the value is 1, repeat the operation on a small area around the point to get a better accuracy. I’ll look into that. For the second problem I’ll add an error message, so we can determine if it actually occurs (biggestSquaredDistanceToBlack is 0). I’ve also thought that having drawn the bitmap, it might be more efficient to then copy its contents out into an array so that finding the distances from pixels inside the polygon to those outside become simple array lookups instead of graphics function calls. I’ll look into that as well.
Cheers,
Mike
*From:*Gerd Petermann [mailto:gpetermann_muenchen@hotmail.com] *Sent:* 07 January 2015 06:55 *To:* mkgmap-dev@lists.mkgmap.org.uk *Subject:* Re: [mkgmap-dev] small issue with Way.getCofG()
Hi Mike,
on my PC the performance of your algo is quite good. Attached is a patch that contains your patch as well as my quick implementation of the algo described here: http://arxiv.org/ftp/arxiv/papers/1212/1212.3193.pdf
The patch tests only performance, it computes the center with the 3 different algos, I've commented the part that prints times and GPX data for debug purposes.
I noticed that the results between both algos are very different, I did not yet try to find out which one is better, but mine is much slower on my PC. I also noticed that your algo doesn't always calculate a point in the polygon, see e.g. way 178708143.
If you like, please try to find a better compromise, I like to fix a problem in splitter first. I also did not yet look at the effect on the house number code, as there are many more small open problems, but I think it should be easy to sort that out.
Gerd
Date: Tue, 6 Jan 2015 13:23:57 -0700 From:gpetermann_muenchen@hotmail.com <mailto:gpetermann_muenchen@hotmail.com> To:mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> Subject: Re: [mkgmap-dev] small issue withWay.getCofG()
Hi Mike,
I like the idea, but it seems to be slow. Is it possible that your algo suffers when no fast graphics card is available? On my netbook the performance is very poor, did not yet try on the PC, but that also has no high speed graphics.
Gerd
GerdP wrote
Hi Mike,
50% sounds better than my algo, but still quite a lot. I'll have a closer look at your algo later. Please note that your change has a side effect on the house number generator. Up to now this doesn't contain a filter for generated POI, so each polygon with a house number is processed twice, once because of the POI, once because the Generator uses Way.getCofG(). If both have different positions this might have a negative impact.
Gerd
From:
mike@.co <mailto:mike@.co>
To:
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
Date: Tue, 6 Jan 2015 14:56:30 +0000 Subject: Re: [mkgmap-dev] small issue withWay.getCofG()
I have a working solution for ensuring that the created point is placed within the polygon when using --add-pois-to-areas, based on drawing the polygon on to a small monochrome bitmap and then looking for the point that is furthest from the surrounding area. I used a 9x9 bitmap for polygons having a small number of points and 15x15 for longer polygons. There is however a performance penalty. My standard map takes about 1 hour 20 minutes; using this algorithm the time increases by about 50% to about 2 hours. I am not currently able to commit changes to SVN (perhaps someone can help out with that) but I have attached the code changes. I suggest that due to the performance penalty, if we adopt this, then the --add-pois-to-areas option be extended to be --add-pois-to-areas[=centre|optimised] or something similar, with the default being centre and functioning as now and the optimised option invoking the new code. Please try out the suggested change. Note I don't expect this to work properly where a polygon is formed from a multiploygon relation, but the code could quite easily be adapted for this circumstance.
Regards, Mike
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev _______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
-- View this message in context:http://gis.19327.n5.nabble.com/small-issue-with-Way-getCofG-tp5828821p582924... Sent from theMkgmap Development mailing list archive at Nabble.com. _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
Hi WanMil, yes, I had something like this in mind. StyledConverter could call the POIGenerator for each polygon. The POIGenerator creates a dummy node, opies the tags, adds the mkgmap:area2poi=true tag and checks if the style produces a map element. If yes, it calculates the coords and returns the node which is than passed to addNode. Anyhow, this change is only useful when performance is a problem. I once had a similar idea for mp-processing. Gerd
Date: Thu, 8 Jan 2015 20:55:46 +0100 From: wmgcnfg@web.de To: mkgmap-dev@lists.mkgmap.org.uk Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
Hi Gerd,
I think there are many drawbacks to move the POI generation into the style code. At least it will complicate the style code.
Maybe it is better to implement your idea directly in the POIGeneratorHook. If it has access to the point style it could check itself if the polygon will (probably) create a POI. Maybe it could first test if there is any overlap of the polygon tags with the tags used in the points file. Then it could pass an artifical POI through the points style and check if a POI is created (finally rules might be disabled to speed up the processing). The hook might also decide how to proceed dependent on the complexity of the polygon.
By the way: Did you think of having different methods depending on the use case? So one getCofG() for POI placement, one getHouseNumberPoint() for housenumber processing etc. It sounds like it is hard to find one algorithm that fits all.
WanMil
Hi Mike,
reg. POI for polygons like rivers etc:
The current algo calculates a POI for each polygon before any style rule is applied. If the calculation of the Coord takes a lot of time, and many of the POI are later dropped in the points rules, maybe it is possible to create a dummy object first, pass it through the style rules, check if it produces a map object, and calculate the exact position only for those. I think we would have to change/remove the POIGeneratorHook and call it "on the fly", but I don't see any big problem with that.
Gerd
------------------------------------------------------------------------ From: mike@tvage.co.uk To: mkgmap-dev@lists.mkgmap.org.uk Date: Wed, 7 Jan 2015 09:29:51 +0000 Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
HI Gerd, the method I’ve used won’t work if the polygon is thinner than the resolution of the bitmap. I hadn’t considered that it might be used on rivers; my main testing has been on car parks, which although they may be irregularly shaped are not normally very long and very thin (I handle river names from the underlying lines). There are two possible problems that could occur – if a polygon is thin enough that it renders in the bitmap as a line that is nowhere wider than 2 pixels the POI could be placed just outside the polygon. I haven’t checked what fillPolygon does if the points have no volume and form a line. If this renders nothing, then the algo will just return the centre point, which could be nowhere near the polygon. A simple solution to the first problem would be to examine the biggestSquaredDistanceToBlack figure before returning the coordinate and if the value is 1, repeat the operation on a small area around the point to get a better accuracy. I’ll look into that. For the second problem I’ll add an error message, so we can determine if it actually occurs (biggestSquaredDistanceToBlack is 0). I’ve also thought that having drawn the bitmap, it might be more efficient to then copy its contents out into an array so that finding the distances from pixels inside the polygon to those outside become simple array lookups instead of graphics function calls. I’ll look into that as well.
Cheers,
Mike
*From:*Gerd Petermann [mailto:gpetermann_muenchen@hotmail.com] *Sent:* 07 January 2015 06:55 *To:* mkgmap-dev@lists.mkgmap.org.uk *Subject:* Re: [mkgmap-dev] small issue with Way.getCofG()
Hi Mike,
on my PC the performance of your algo is quite good. Attached is a patch that contains your patch as well as my quick implementation of the algo described here: http://arxiv.org/ftp/arxiv/papers/1212/1212.3193.pdf
The patch tests only performance, it computes the center with the 3 different algos, I've commented the part that prints times and GPX data for debug purposes.
I noticed that the results between both algos are very different, I did not yet try to find out which one is better, but mine is much slower on my PC. I also noticed that your algo doesn't always calculate a point in the polygon, see e.g. way 178708143.
If you like, please try to find a better compromise, I like to fix a problem in splitter first. I also did not yet look at the effect on the house number code, as there are many more small open problems, but I think it should be easy to sort that out.
Gerd
Date: Tue, 6 Jan 2015 13:23:57 -0700 From:gpetermann_muenchen@hotmail.com <mailto:gpetermann_muenchen@hotmail.com> To:mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> Subject: Re: [mkgmap-dev] small issue withWay.getCofG()
Hi Mike,
I like the idea, but it seems to be slow. Is it possible that your algo suffers when no fast graphics card is available? On my netbook the performance is very poor, did not yet try on the PC, but that also has no high speed graphics.
Gerd
GerdP wrote
Hi Mike,
50% sounds better than my algo, but still quite a lot. I'll have a closer look at your algo later. Please note that your change has a side effect on the house number generator. Up to now this doesn't contain a filter for generated POI, so each polygon with a house number is processed twice, once because of the POI, once because the Generator uses Way.getCofG(). If both have different positions this might have a negative impact.
Gerd
From:
mike@.co <mailto:mike@.co>
To:
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
Date: Tue, 6 Jan 2015 14:56:30 +0000 Subject: Re: [mkgmap-dev] small issue withWay.getCofG()
I have a working solution for ensuring that the created point is placed within the polygon when using --add-pois-to-areas, based on drawing the polygon on to a small monochrome bitmap and then looking for the point that is furthest from the surrounding area. I used a 9x9 bitmap for polygons having a small number of points and 15x15 for longer polygons. There is however a performance penalty. My standard map takes about 1 hour 20 minutes; using this algorithm the time increases by about 50% to about 2 hours. I am not currently able to commit changes to SVN (perhaps someone can help out with that) but I have attached the code changes. I suggest that due to the performance penalty, if we adopt this, then the --add-pois-to-areas option be extended to be --add-pois-to-areas[=centre|optimised] or something similar, with the default being centre and functioning as now and the optimised option invoking the new code. Please try out the suggested change. Note I don't expect this to work properly where a polygon is formed from a multiploygon relation, but the code could quite easily be adapted for this circumstance.
Regards, Mike
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev _______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
-- View this message in context:http://gis.19327.n5.nabble.com/small-issue-with-Way-getCofG-tp5828821p582924... Sent from theMkgmap Development mailing list archive at Nabble.com. _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
Hi Gerd/WanMil, I now have a mechanism working that has no performance overhead. I have derived a new class from Node and have moved the evaluation of the location to when getLocation is first called in the new class. I have also improved the algorithm by progressively using bigger bitmaps if a small one doesn't produce a suitable result. I'm now looking at doing the same for multi-polygons before putting together a patch. I should have something for you to try shortly. Regards, Mike From: Gerd Petermann [mailto:gpetermann_muenchen@hotmail.com] Sent: 08 January 2015 20:23 To: mkgmap-dev@lists.mkgmap.org.uk Subject: Re: [mkgmap-dev] small issue with Way.getCofG() Hi WanMil, yes, I had something like this in mind. StyledConverter could call the POIGenerator for each polygon. The POIGenerator creates a dummy node, opies the tags, adds the mkgmap:area2poi=true tag and checks if the style produces a map element. If yes, it calculates the coords and returns the node which is than passed to addNode. Anyhow, this change is only useful when performance is a problem. I once had a similar idea for mp-processing. Gerd
Date: Thu, 8 Jan 2015 20:55:46 +0100 From: wmgcnfg@web.de <mailto:wmgcnfg@web.de> To: mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk>
Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
Hi Gerd,
I think there are many drawbacks to move the POI generation into the style code. At least it will complicate the style code.
Maybe it is better to implement your idea directly in the POIGeneratorHook. If it has access to the point style it could check itself if the polygon will (probably) create a POI. Maybe it could first test if there is any overlap of the polygon tags with the tags used in the points file. Then it could pass an artifical POI through the points style and check if a POI is created (finally rules might be disabled to speed up the processing). The hook might also decide how to proceed dependent on the complexity of the polygon.
By the way: Did you think of having different methods depending on the use case? So one getCofG() for POI placement, one getHouseNumberPoint() for housenumber processing etc. It sounds like it is hard to find one algorithm that fits all.
WanMil
Hi Mike,
reg. POI for polygons like rivers etc:
The current algo calculates a POI for each polygon before any style rule is applied. If the calculation of the Coord takes a lot of time, and many of the POI are later dropped in the points rules, maybe it is possible to create a dummy object first, pass it through the style rules, check if it produces a map object, and calculate the exact position only for those. I think we would have to change/remove the POIGeneratorHook and call it "on the fly", but I don't see any big problem with that.
Gerd
------------------------------------------------------------------------ From: mike@tvage.co.uk <mailto:mike@tvage.co.uk> To: mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> Date: Wed, 7 Jan 2015 09:29:51 +0000 Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
HI Gerd, the method I've used won't work if the polygon is thinner than the resolution of the bitmap. I hadn't considered that it might be used on rivers; my main testing has been on car parks, which although they may be irregularly shaped are not normally very long and very thin (I handle river names from the underlying lines). There are two possible problems that could occur - if a polygon is thin enough that it renders in the bitmap as a line that is nowhere wider than 2 pixels the POI could be placed just outside the polygon. I haven't checked what fillPolygon does if the points have no volume and form a line. If this renders nothing, then the algo will just return the centre point, which could be nowhere near the polygon. A simple solution to the first problem would be to examine the biggestSquaredDistanceToBlack figure before returning the coordinate and if the value is 1, repeat the operation on a small area around the point to get a better accuracy. I'll look into that. For the second problem I'll add an error message, so we can determine if it actually occurs (biggestSquaredDistanceToBlack is 0). I've also thought that having drawn the bitmap, it might be more efficient to then copy its contents out into an array so that finding the distances from pixels inside the polygon to those outside become simple array lookups instead of graphics function calls. I'll look into that as well.
Cheers,
Mike
*From:*Gerd Petermann [mailto:gpetermann_muenchen@hotmail.com] *Sent:* 07 January 2015 06:55 *To:* mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> *Subject:* Re: [mkgmap-dev] small issue with Way.getCofG()
Hi Mike,
on my PC the performance of your algo is quite good. Attached is a patch that contains your patch as well as my quick implementation of the algo described here: http://arxiv.org/ftp/arxiv/papers/1212/1212.3193.pdf
The patch tests only performance, it computes the center with the 3 different algos, I've commented the part that prints times and GPX data for debug purposes.
I noticed that the results between both algos are very different, I did not yet try to find out which one is better, but mine is much slower on my PC. I also noticed that your algo doesn't always calculate a point in the polygon, see e.g. way 178708143.
If you like, please try to find a better compromise, I like to fix a problem in splitter first. I also did not yet look at the effect on the house number code, as there are many more small open problems, but I think it should be easy to sort that out.
Gerd
Date: Tue, 6 Jan 2015 13:23:57 -0700 From:gpetermann_muenchen@hotmail.com <mailto:gpetermann_muenchen@hotmail.com> To:mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> Subject: Re: [mkgmap-dev] small issue withWay.getCofG()
Hi Mike,
I like the idea, but it seems to be slow. Is it possible that your algo suffers when no fast graphics card is available? On my netbook the performance is very poor, did not yet try on the PC, but that also has no high speed graphics.
Gerd
GerdP wrote
Hi Mike,
50% sounds better than my algo, but still quite a lot. I'll have a closer look at your algo later. Please note that your change has a side effect on the house number generator. Up to now this doesn't contain a filter for generated POI, so each polygon with a house number is processed twice, once because of the POI, once because the Generator uses Way.getCofG(). If both have different positions this might have a negative impact.
Gerd
From:
mike@.co <mailto:mike@.co> <mailto:mike@.co>
To:
mkgmap-dev@.org <mailto:mkgmap-dev@.org> <mailto:mkgmap-dev@.org>
Date: Tue, 6 Jan 2015 14:56:30 +0000 Subject: Re: [mkgmap-dev] small issue withWay.getCofG()
I have a working solution for ensuring that the created point is placed within the polygon when using --add-pois-to-areas, based on drawing the polygon on to a small monochrome bitmap and then looking for the point that is furthest from the surrounding area. I used a 9x9 bitmap for polygons having a small number of points and 15x15 for longer polygons. There is however a performance penalty. My standard map takes about 1 hour 20 minutes; using this algorithm the time increases by about 50% to about 2 hours. I am not currently able to commit changes to SVN (perhaps someone can help out with that) but I have attached the code changes. I suggest that due to the performance penalty, if we adopt this, then the --add-pois-to-areas option be extended to be --add-pois-to-areas[=centre|optimised] or something similar, with the default being centre and functioning as now and the optimised option invoking the new code. Please try out the suggested change. Note I don't expect this to work properly where a polygon is formed from a multiploygon relation, but the code could quite easily be adapted for this circumstance.
Regards, Mike
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org <mailto:mkgmap-dev@.org> <mailto:mkgmap-dev@.org>
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev _______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org <mailto:mkgmap-dev@.org> <mailto:mkgmap-dev@.org>
-- View this message in context:http://gis.19327.n5.nabble.com/small-issue-with-Way-getCofG-tp582882 1p5829247.html Sent from theMkgmap Development mailing list archive at Nabble.com. _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> <mailto:mkgmap-dev@lists.mkgmap.org.uk> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
Hi Miike, sounds great! Reg. mp-processing: Not sure what you plan to do here. Maybe that is not so easy, keep in mind that we have the area_size() style function, so you probably have to calculate at least the outer shape before you can check if the style is creating a map object. Gerd From: mike@tvage.co.uk To: mkgmap-dev@lists.mkgmap.org.uk Date: Fri, 9 Jan 2015 00:20:56 +0000 Subject: Re: [mkgmap-dev] small issue with Way.getCofG() Hi Gerd/WanMil, I now have a mechanism working that has no performance overhead. I have derived a new class from Node and have moved the evaluation of the location to when getLocation is first called in the new class. I have also improved the algorithm by progressively using bigger bitmaps if a small one doesn’t produce a suitable result. I’m now looking at doing the same for multi-polygons before putting together a patch. I should have something for you to try shortly. Regards,Mike From: Gerd Petermann [mailto:gpetermann_muenchen@hotmail.com] Sent: 08 January 2015 20:23 To: mkgmap-dev@lists.mkgmap.org.uk Subject: Re: [mkgmap-dev] small issue with Way.getCofG() Hi WanMil, yes, I had something like this in mind. StyledConverter could call the POIGenerator for each polygon. The POIGenerator creates a dummy node, opies the tags, adds the mkgmap:area2poi=true tag and checks if the style produces a map element. If yes, it calculates the coords and returns the node which is than passed to addNode. Anyhow, this change is only useful when performance is a problem. I once had a similar idea for mp-processing. Gerd> Date: Thu, 8 Jan 2015 20:55:46 +0100
From: wmgcnfg@web.de To: mkgmap-dev@lists.mkgmap.org.uk Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
Hi Gerd,
I think there are many drawbacks to move the POI generation into the style code. At least it will complicate the style code.
Maybe it is better to implement your idea directly in the POIGeneratorHook. If it has access to the point style it could check itself if the polygon will (probably) create a POI. Maybe it could first test if there is any overlap of the polygon tags with the tags used in the points file. Then it could pass an artifical POI through the points style and check if a POI is created (finally rules might be disabled to speed up the processing). The hook might also decide how to proceed dependent on the complexity of the polygon.
By the way: Did you think of having different methods depending on the use case? So one getCofG() for POI placement, one getHouseNumberPoint() for housenumber processing etc. It sounds like it is hard to find one algorithm that fits all.
WanMil
Hi Mike,
reg. POI for polygons like rivers etc:
The current algo calculates a POI for each polygon before any style rule is applied. If the calculation of the Coord takes a lot of time, and many of the POI are later dropped in the points rules, maybe it is possible to create a dummy object first, pass it through the style rules, check if it produces a map object, and calculate the exact position only for those. I think we would have to change/remove the POIGeneratorHook and call it "on the fly", but I don't see any big problem with that.
Gerd
------------------------------------------------------------------------ From: mike@tvage.co.uk To: mkgmap-dev@lists.mkgmap.org.uk Date: Wed, 7 Jan 2015 09:29:51 +0000 Subject: Re: [mkgmap-dev] small issue with Way.getCofG()
HI Gerd, the method I’ve used won’t work if the polygon is thinner than the resolution of the bitmap. I hadn’t considered that it might be used on rivers; my main testing has been on car parks, which although they may be irregularly shaped are not normally very long and very thin (I handle river names from the underlying lines). There are two possible problems that could occur – if a polygon is thin enough that it renders in the bitmap as a line that is nowhere wider than 2 pixels the POI could be placed just outside the polygon. I haven’t checked what fillPolygon does if the points have no volume and form a line. If this renders nothing, then the algo will just return the centre point, which could be nowhere near the polygon. A simple solution to the first problem would be to examine the biggestSquaredDistanceToBlack figure before returning the coordinate and if the value is 1, repeat the operation on a small area around the point to get a better accuracy. I’ll look into that. For the second problem I’ll add an error message, so we can determine if it actually occurs (biggestSquaredDistanceToBlack is 0). I’ve also thought that having drawn the bitmap, it might be more efficient to then copy its contents out into an array so that finding the distances from pixels inside the polygon to those outside become simple array lookups instead of graphics function calls. I’ll look into that as well.
Cheers,
Mike
*From:*Gerd Petermann [mailto:gpetermann_muenchen@hotmail.com] *Sent:* 07 January 2015 06:55 *To:* mkgmap-dev@lists.mkgmap.org.uk *Subject:* Re: [mkgmap-dev] small issue with Way.getCofG()
Hi Mike,
on my PC the performance of your algo is quite good. Attached is a patch that contains your patch as well as my quick implementation of the algo described here: http://arxiv.org/ftp/arxiv/papers/1212/1212.3193.pdf
The patch tests only performance, it computes the center with the 3 different algos, I've commented the part that prints times and GPX data for debug purposes.
I noticed that the results between both algos are very different, I did not yet try to find out which one is better, but mine is much slower on my PC. I also noticed that your algo doesn't always calculate a point in the polygon, see e.g. way 178708143.
If you like, please try to find a better compromise, I like to fix a problem in splitter first. I also did not yet look at the effect on the house number code, as there are many more small open problems, but I think it should be easy to sort that out.
Gerd
Date: Tue, 6 Jan 2015 13:23:57 -0700 From:gpetermann_muenchen@hotmail.com <mailto:gpetermann_muenchen@hotmail.com> To:mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> Subject: Re: [mkgmap-dev] small issue withWay.getCofG()
Hi Mike,
I like the idea, but it seems to be slow. Is it possible that your algo suffers when no fast graphics card is available? On my netbook the performance is very poor, did not yet try on the PC, but that also has no high speed graphics.
Gerd
GerdP wrote
Hi Mike,
50% sounds better than my algo, but still quite a lot. I'll have a closer look at your algo later. Please note that your change has a side effect on the house number generator. Up to now this doesn't contain a filter for generated POI, so each polygon with a house number is processed twice, once because of the POI, once because the Generator uses Way.getCofG(). If both have different positions this might have a negative impact.
Gerd
From:
mike@.co <mailto:mike@.co>
To:
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
Date: Tue, 6 Jan 2015 14:56:30 +0000 Subject: Re: [mkgmap-dev] small issue withWay.getCofG()
I have a working solution for ensuring that the created point is placed within the polygon when using --add-pois-to-areas, based on drawing the polygon on to a small monochrome bitmap and then looking for the point that is furthest from the surrounding area. I used a 9x9 bitmap for polygons having a small number of points and 15x15 for longer polygons. There is however a performance penalty. My standard map takes about 1 hour 20 minutes; using this algorithm the time increases by about 50% to about 2 hours. I am not currently able to commit changes to SVN (perhaps someone can help out with that) but I have attached the code changes. I suggest that due to the performance penalty, if we adopt this, then the --add-pois-to-areas option be extended to be --add-pois-to-areas[=centre|optimised] or something similar, with the default being centre and functioning as now and the optimised option invoking the new code. Please try out the suggested change. Note I don't expect this to work properly where a polygon is formed from a multiploygon relation, but the code could quite easily be adapted for this circumstance.
Regards, Mike
_______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev _______________________________________________ mkgmap-dev mailing list
mkgmap-dev@.org <mailto:mkgmap-dev@.org>
-- View this message in context:http://gis.19327.n5.nabble.com/small-issue-with-Way-getCofG-tp5828821p582924... Sent from theMkgmap Development mailing list archive at Nabble.com. _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk <mailto:mkgmap-dev@lists.mkgmap.org.uk> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
participants (4)
-
Gerd Petermann -
GerdP
-
Mike Baggaley -
WanMil