Nous avons un sujet en français, voir ici http://www.x-plane.fr/thread49914.html
We already have a subject in french, see link above.
As Andras Fabian (Alpilotx) does not understand french, I created this subject here in the english speaking part of the french forum.
Here the mail I received from him:
Feel free to ask questions here,
Daniel
We already have a subject in french, see link above.
As Andras Fabian (Alpilotx) does not understand french, I created this subject here in the english speaking part of the french forum.
Here the mail I received from him:
Hi Daniel!
Nice to hear back from you. And, well, I was not really following the
conversation ... but at least tried to figure out, what might be in
there (and I have seen, that obviously there were "questions").
To start with, I would highly recommend everybody who can read English,
to check out this interview which I did with xplane10s Blog in December:
It covers quite a bit of the background details about the base scenery!
...
There was also an important change between XP9 and XP10. In XP9, the
triangle mesh of the terrain was a 3D mesh. Which means, the mesh nodes
had Z (elevation) information too, and added vertex normals. With XP10,
the triangle mesh is only saved in 2D. Like a flat triangle mesh! And in
the DSF we have now the "original" elevation raster data saved too. So,
what XP10 does is, that it takes the 2D triangle mesh - like a veil -
and puts it over the elevation raster data, and lets it take that shape.
For one, this saves a little space in th DSFs (as it compresses much
better this way, and needs a bit less "redundant" information). AND
there is some other "secret" (well, Ben blogged about this too): this
information can be used later, to reallyrefine the triangle mesh itself
(which is not possible until now): its called "hardware tessellation"! A
technique to refine triangle structures at runtime (by braking them up
in smaller triangles) on the 3D hardware. And of course, with that
refined mesh, the landscape could more accurately mimic the raster data.
And no, this is no real secret :-) Ben blogged about this here:
Then, something about the raw data. SRTM is not incorrect, but neither
entirely correct :-) . Yes, we use it in many places BUT there are also
other sources. For all the northern latitudes (above 60 degrees North,
there is no SRTM data!), and for many places where available, we use the
fantastic data from http://www.viewfinderpanoramas.org/ (the 90m
resolution data)! Yes, in the Alps too (and in the Alps,
www.viewfinderpanoramas.org would even have 30m res data). Its a really
good source (I love it, as it is - even at the same 90m resolution -
more accurate than SRTM ... I have a few examples in the Dolomites,
where I can really see the difference)!
BUT still, the resolution of the triangle mesh structure is the
limiting factor! So, if the triangles are too big, then the high res
raster elevation data does not help much (at least not until we have
hardware tessellation - somewhere in the future, don't ask me when). And
that is where my HD Mesh tries to improve. More and smaller triangles
result in better representation of the raster elevation data.
By the way, there is also another advantage of the HD Mesh. Not only
elevation information is bound to the triangle mesh, but also landclass
information. When the DSFs are created, our processing tool assigns a
terrain type (indirectly - via library.txt - these are the TER files)
toeach triangle! And our Landclass data has much higher resolution
(around 100m in Europe and USA) than the triangle mesh of the default
Global Scenery can depict. Here, again, smaller triangles help to have a
more exact representation of landclass information too (AND this is a
poitn, which can't be improved by hardware tessellation - it can#t
assign new terrain types on the fly to the smaller triangles ... at
least not in the next years).
Finally, here is a low-level screenshot from the GUI of our scenery
creator (called RenderFarm), which compares the default triangle mesh
(as its is in the default Global Scenery) with the HD Mesh. I think,
here, everybody can see the difference:
I hope I could answer a few questions ... and maybe, if your time
permits, you might pass this on in your forum (if someone is interested
in it at all :-) )!
Have a nice day,
Andras
Nice to hear back from you. And, well, I was not really following the
conversation ... but at least tried to figure out, what might be in
there (and I have seen, that obviously there were "questions").
To start with, I would highly recommend everybody who can read English,
to check out this interview which I did with xplane10s Blog in December:
It covers quite a bit of the background details about the base scenery!
...
There was also an important change between XP9 and XP10. In XP9, the
triangle mesh of the terrain was a 3D mesh. Which means, the mesh nodes
had Z (elevation) information too, and added vertex normals. With XP10,
the triangle mesh is only saved in 2D. Like a flat triangle mesh! And in
the DSF we have now the "original" elevation raster data saved too. So,
what XP10 does is, that it takes the 2D triangle mesh - like a veil -
and puts it over the elevation raster data, and lets it take that shape.
For one, this saves a little space in th DSFs (as it compresses much
better this way, and needs a bit less "redundant" information). AND
there is some other "secret" (well, Ben blogged about this too): this
information can be used later, to reallyrefine the triangle mesh itself
(which is not possible until now): its called "hardware tessellation"! A
technique to refine triangle structures at runtime (by braking them up
in smaller triangles) on the 3D hardware. And of course, with that
refined mesh, the landscape could more accurately mimic the raster data.
And no, this is no real secret :-) Ben blogged about this here:
Then, something about the raw data. SRTM is not incorrect, but neither
entirely correct :-) . Yes, we use it in many places BUT there are also
other sources. For all the northern latitudes (above 60 degrees North,
there is no SRTM data!), and for many places where available, we use the
fantastic data from http://www.viewfinderpanoramas.org/ (the 90m
resolution data)! Yes, in the Alps too (and in the Alps,
www.viewfinderpanoramas.org would even have 30m res data). Its a really
good source (I love it, as it is - even at the same 90m resolution -
more accurate than SRTM ... I have a few examples in the Dolomites,
where I can really see the difference)!
BUT still, the resolution of the triangle mesh structure is the
limiting factor! So, if the triangles are too big, then the high res
raster elevation data does not help much (at least not until we have
hardware tessellation - somewhere in the future, don't ask me when). And
that is where my HD Mesh tries to improve. More and smaller triangles
result in better representation of the raster elevation data.
By the way, there is also another advantage of the HD Mesh. Not only
elevation information is bound to the triangle mesh, but also landclass
information. When the DSFs are created, our processing tool assigns a
terrain type (indirectly - via library.txt - these are the TER files)
toeach triangle! And our Landclass data has much higher resolution
(around 100m in Europe and USA) than the triangle mesh of the default
Global Scenery can depict. Here, again, smaller triangles help to have a
more exact representation of landclass information too (AND this is a
poitn, which can't be improved by hardware tessellation - it can#t
assign new terrain types on the fly to the smaller triangles ... at
least not in the next years).
Finally, here is a low-level screenshot from the GUI of our scenery
creator (called RenderFarm), which compares the default triangle mesh
(as its is in the default Global Scenery) with the HD Mesh. I think,
here, everybody can see the difference:
I hope I could answer a few questions ... and maybe, if your time
permits, you might pass this on in your forum (if someone is interested
in it at all :-) )!
Have a nice day,
Andras
Daniel
Commentaire