New Celestia-1.4.0pre-FT1.1 Version for Download

[Toti]

Hi guys,

There are a number of reasons why the output of that program is so unreliable:

--I spent lots of my spare time trying to find an appropriate set of parametric equations to generate all Hubble types with the appropiate density curves, etc. To find the expressions of the best generators I had to play with tricky integrals to ensure C2 continuity, etc. which ended up being either annalitycally unsolvable or led to unusable functions; so I just gave up and switched to an image based approach. I hacked the bmp2pts tool to just "test the terrain" with the purpose of enhace it later. Most of this can be found in this thread

--Celestia needs a very special size distribution of the textured kernels (the "blobs") to work properly. Needless to say, no galaxy image follows this rule, so one has to make some extreme simplification assumptions. Also the number of blobs must be minimized somehow.


On a related topic, the whole system is designed to give the best results when watched from far away. This is because Celestia must run in a wide range of video hardware: whilst some people can afford a FX6800, many others have much older and less capable cards, but the program must run as fast as possible also for them. The blending speed of the system is crucial here and most hardware is simply not fast enough, so some trimming must be done. This is the reason why galaxies lose their gaseous feel when viewed at close range.


Finally, I can assure you that Fridger really put a lot of effort on those templates, and worked on them for several weeks! I really doubt they can be much better than this. If someone prefers an older one then it is just as easy as replacing the corresponding .pts file.

[t00fri]

I should add that beyond what you can see, there was a lot of /quantitative/ image manipulation work behind EACH /input/ .bmp file from which the templates were made:

-- They must be precisely scaled within the 128x128 image. Otherwise the resulting galactic extensions become incorrect!

-- They must be rotated to a reference position by means of GIMP
Otherwise the subsequent rotations will do the wrong thing.

-- They must be unfolded by projection to correspond to face-on view! Otherwise the inclination and position angle from the catalog will generate the wrong orientation!

-- They must be relatively normalized in absolute brightness values, otherwise, you loose the /relative/ apparent brightness feature of the displayed galaxies.

NONE of the earlier galaxy models had any of these crucial tasks implemented. Hence for any more ambitious final setup, one has to do all these time-consuming tasks first.

I guess you can see that there is a lot of tricky work to be done before you can even think about the final template image quality. In addition, when you modiify the brightness to affect the .pts point distribution, the visual galaxy extension usually changes and one has to rescale.

Bye Fridger

[t00fri]

Typically a .pts file has 5000 such points or more.

I believe the current template for Sb galaxies has 9,793 points.

- Hank

Which is correct and in no contradiction with my statement. But LESS than 5000 points are usually unusable. 5000 points typically correspond to ~200KB file size. Since we shall switch to binary format soon, we may save a factor of two here. But the better templates all have between 7000-9000 points, right now.

[abramson]

These .pts files can be displayed in Celestia or other software like Maple, but NOT edited anymore! Typically a .pts file has 5000 such points or more.

Why is this, Fridger or Toti? After reading the messages of the last days I have been playing with sbc.pts, applying statistical transformations to see what happens. What happens is always uglier, I have to add. (I didn't have much time to play, also.)

Is it because of the authors (F & T) don't want fiddling with these files? In that case I will stop doing it immediately. Please let me know.

Regards,

Guillermo

PS: Let me add that I like a little asymmetry in the galaxies, I believe that it gives them more realism.

[ElChristou]

These .pts files can be displayed in Celestia or other software like Maple, but NOT edited anymore! Typically a .pts file has 5000 such points or more.

Why is this, Fridger or Toti? After reading the messages of the last days I have been playing with sbc.pts, applying statistical transformations to see what happens. What happens is always uglier, I have to add. (I didn't have much time to play, also.)

Is it because of the authors (F & T) don't want fiddling with these files? In that case I will stop doing it immediately. Please let me know.

I think Fridger want to say that the file cannot be edited anymore in the meaning that you won't have an easy control of what you are doing... finding some specific points in a .pts file and changing his position and brightness is a really hard task I suppose...

[t00fri]

These .pts files can be displayed in Celestia or other software like Maple, but NOT edited anymore! Typically a .pts file has 5000 such points or more.

Why is this, Fridger or Toti? After reading the messages of the last days I have been playing with sbc.pts, applying statistical transformations to see what happens. What happens is always uglier, I have to add. (I didn't have much time to play, also.)

Is it because of the authors (F & T) don't want fiddling with these files? In that case I will stop doing it immediately. Please let me know.

Regards,

Guillermo

PS: Let me add that I like a little asymmetry in the galaxies, I believe that it gives them more realism.

Guillermo,

sure enough can one apply mathematical transformations to this 4-component vector of data. But editing does NOT work in the sense one normally does with image manipulation programs (interactively). Sure you may also take a few points out by hand. But given 7000-9000 points on average, that's a bit frustrating in the long run

We don't mind at all if people want to fiddle around with the .pts files. To the contrary. Perhaps someone gets a good idea this way ...

Cheers,
Fridger

[abramson]

We don't mind at all if people want to fiddle around with the .pts files. To the contrary. Perhaps someone gets a good idea this way ...

Ok (phew!). Until now, as I said, nothing nice surfaced, and given your experience (for example what Toti says in his recent post about trying to model the galaxy types parametrically), perhaps nothing will. If something comes up, I will post it.

Guillermo

[hank]

Typically a .pts file has 5000 such points or more.

I believe the current template for Sb galaxies has 9,793 points.

- Hank

Which is correct and in no contradiction with my statement. But LESS than 5000 points are usually unusable. 5000 points typically correspond to ~200KB file size. Since we shall switch to binary format soon, we may save a factor of two here. But the better templates all have between 7000-9000 points, right now.

Right. I did not intend to contradict your statement, but rather to emphasize that there are LOTS of points.

But editing does NOT work in the sense one normally does with image manipulation programs (interactively). Sure you may also take a few points out by hand. But given 7000-9000 points on average, that's a bit frustrating in the long run ;-)

Which was my intended point, as well.

Given this fact, how might an interactive blob editor work? Anyone have any ideas?

- Hank

[hank]

So thinking about .bmp template images without trying them out right away through Toti's 'bmp2pts' is COMPLETELY useless!

Toti's code is standard C++ code that you may compile in any OS you like. Toti is just (correctly) reluctant to distribute it, since then we get so many questions that we have to stop working or ignore the questions. It only makes sense to distribute it to people who can read directly from the code what the program is doing and HOW it is doing it.

Would it be possible to convert Toti's 'bmp2pts' to Lua? How much code is there? Does it use a lot of fancy c++ stuff?

- Hank

[t00fri]

So thinking about .bmp template images without trying them out right away through Toti's 'bmp2pts' is COMPLETELY useless!

Toti's code is standard C++ code that you may compile in any OS you like. Toti is just (correctly) reluctant to distribute it, since then we get so many questions that we have to stop working or ignore the questions. It only makes sense to distribute it to people who can read directly from the code what the program is doing and HOW it is doing it.

Would it be possible to convert Toti's 'bmp2pts' to Lua? How much code is there? Does it use a lot of fancy c++ stuff?

- Hank

The code is not fancy, but I don't see what a a conversion to Lua should really gain over a clever script for automatizing all these repeated steps?

-- modify .bmp with GIMP, PS
-- call bmp2pts with suitable parameters.
If # points < 5000, lower the threshold, call again...etc
--copy to /usr/local/share/celestia/models
--call celestia
--select and display a particular galaxy for test
-- go to beginning, repeat...

ElChristou, is already close to desperation because he has convinced himself meanwhile in intensive email interaction with me that this template business is HORROR. I converted his tries in bmp2pts every time, trying to make the best of his attempts...

The problem is this: thinking and trying to be systematic makes things worse! bmp2pts has its own will that is almost impossible to break . It's a bit like learning to play 5 balls with 2 hands at a wall. You have to practice a LONG time and one day you can somehow manage, without really knowing why and how ...

++++++++++++++
I think we should next ask Cham to try his luck, since he was the loudest criticizer of the present quality of the templates
++++++++++++++
Cheers,
Bye Fridger

[t00fri]

OK, within the little time I have for the next few weeks, I volounteer to convert people's tries, sending the .pts files back.

The options are these

Code: Select all

bmp2pts <bmp-file> <threshold> <randomness>


The threshold value determines the sensitivity of the program to assign a dot to a small region in the .bmp input.

The randomness just washes out the result a bit with sensible values ranging between 0.03 and 0.05

The threshold should be lowered until at least 5000 points are generated, better 7000-9000.

That's about it

Bye Fridger

PS: Remember grayscale 128x128 bmp input required.

[ElChristou]

Fridger is right, indeed Toti's tool react in really strange ways... I will do some tries, but the result won't be (or only a bit ) better than Fridger's ones...

[t00fri]

Well perhaps Toti can be convinced to make his tool available for people to practice (after its compilation). I don't want to interfer. It's Toti's code (I just found a few bugs) . Certainly questions referring to the tool should be forbidden in case it's made available.

Toti?

Bye Fridger

PS: If you get a SEGFAULT just lower the threshold a bit.
The code segfaults if the number of generated points is <0

[t00fri]

... but the result won't be (or only a bit ) better than Fridger's ones...

Hey hey! NEVER

Cheers,
Fridger

[hank]

The code is not fancy, but I don't see what a a conversion to Lua should really gain over a clever script for automatizing all these repeated steps?

-- modify .bmp with GIMP, PS
-- call bmp2pts with suitable parameters.
If # points < 5000, lower the threshold, call again...etc
--copy to /usr/local/share/celestia/models
--call celestia
--select and display a particular galaxy for test
-- go to beginning, repeat...

The advantage would be that if implemented in Lua it could be called directly from a CELX script. This would possibly allow immediate updating of the galaxy displayed in Celestia. Then the process would be something like:

start celestia
select and display a particular galaxy for test
run the CELX script
repeat
...modify .bmp with GIMP, PS
...press a key in Celestia to have the script update the galaxy template
until the galaxy looks right
press a key in Celestia to have the script save the new galaxy template

I suppose the same thing could be done having Lua call the os to run bmp2pts (at least on Linux and Mac OS X), but having it all in Lua seems cleaner. Plus, I really like Lua.

- Hank

[t00fri]

Hank,

would the template be updated without restarting Celestia?
I am not sure.

Fridger

[Toti]

Given this fact, how might an interactive blob editor work? Anyone have any ideas?

I think it would be a nightmare. It's easy to code it in Blender's Python interface, but the "wireframe" of such a mess of 5000-9000 squared blobs will be extremely difficult to edit interactively. Also the final result has a strong dependence in the resulting overlap of blobs, which is also hard to control.

Well perhaps Toti can be convinced to make his tool available for people to practice (after its compilation). I don't want to interfer. It's Toti's code (I just found a few bugs) . Certainly questions referring to the tool should be forbidden in case it's made available.

Although I don't mind people playing with it, I don't think it is a good idea: the new tool is taking shape and its results are uncomparably better.

[ElChristou]

Although I don't mind people playing with it, I don't think it is a good idea: the new tool is taking shape and its results are uncomparably better.

Very good news... but indeed using the actual is a very good challenge!!

[hank]

Hank,

would the template be updated without restarting Celestia?
I am not sure.

Fridger

You would actually be modifying the internal GalacticForm data directly rather than the template file. The template file would only be updated to save the result. Provided you have access via CELX, I believe this would work.

- Hank

[fsgregs]

Fridger:

Since you are going to be busy for the next 10 days or so and out of the country, I am (GULP! ) going to ask (Don't get upset) ... if you could post the edited FT 1.1 file that fixes the flickering nebula problem? It can be just a Windows file ... nothing fancy. You already have it done ... and it seems the number of bug reports coming in on FT 1.1 have dropped to background level ... could this qualify as a reasonable time to release FT 1.1.1



Frank