xNormal - MASTER THREAD
polycounter lvl 20
just saw this on opengl.org, might be worth a try, specially with the seemingly not so well working melody
didnt have time to test it, but a well known modeller has helped development
http://www.xnormal.net
didnt have time to test it, but a well known modeller has helped development
http://www.xnormal.net
Replies
[Email]jogshy [at] mixmail.com[/Email]
That strategery worked great for me with Kaldera, I got a great tool out of it.
I'm taking good note about all your feedback. Atm I am finishing the quad support implementation as well as solving some other bugs.
About the close button... There is! Is that ugly small blue button in the bottom-left! I know my UI art sux badly but... LOL...
I dont know what is the floating geometry thing... Can anybody explain to me pls?
I really wanna thank you all the feedback these days I am getting to improve the program. Please if you have suggestions, bugs or whatever, you can contact me at [Email]jogshy [at] mixmail.com[/Email] as usually or in my web page.
http://www.iddevnet.com/quake4/ArtReference_CreatingModels
Scroll down until you see the picture of the hi-res knee with the triangular indent.
Hope you like it. thx for all the feedback.
Thanks again!
Gonna test it asap!
The normal-map renderer works fine, very fast and quite a good normal map output after tweaking the ray cast distance. However when I tried to turn on the Ambient Occlusion render too, the program would crash every time (usually halfway through the "mipmap 0 level generation" process).
Is this a known bug? My highpoly had floating geometry and was all quads, the lowpoly was all triangles. I was generating a 512x512 texture.
Edit: Hmm, no, the program crashes at the same point (calculating normals, [mip level 0], just after halfway on the progress bar) with version 3.1.1.
However you were right about the dual monitor thing I emailed you about, I had 2 -> 1 instead of 1 -> 2
Well, expect a new version in a few days once I locate the problem.
Btw, I added in my web a small .txt with the changes, so you no longer need to download the 11Mb to see them hehe...
Basically, solved the ambient occlusion crash ( was a problem with Pentium4 dual-cores and hyperthreading ), added camera orbit mode, added Stanford PLY mesh format ( because it appears is very popular in 3d-scanners ), now the 3d viewer supports floating point textures for improved normal map appereance as well as FSAA and anisotropic filtering.
You can download it as usually in http://www.santyesprogramadorynografista.net/projects.aspx
Hope you like it!
So I just installed your new version and Im still getting a crash at the obj reading stage the same as the prior version. Im wondering If you have a limit either hardcoded or memory based on polycount? My hi poly is humungous ( approx 750,000 tris ) and I'm wondering If this is related to the crash? Any thoughts appreciated. Btw, this is without the AO pass ticked. I have 1.5 gig of Ram If that info helps any. I'd send you the obj but its 75 meg ;-p
Also I think the 3d viewer is an awesome tool, since some programs don't have a previewer for this sort of thing, and often I've seen people asking where they can get a free program to view lowpoly meshes with normal maps applied - in future I will tell them to go and use your program
I took a screenshot of one of my models in it, with diffuse, normal, emissive, spec. power, colour and glossiness maps, and it works very well! You can see a reduced size screengrab here: http://ldaustinart.com/paul/xnormal.jpg
I noticed that my specular/glossiness maps look different in this viewer than in 3dsmax, do you know why that might be? The size of the highlight seems much smaller than the glossiness map should make it seem... any idea why this might be? Is it something I'm doing wrong, or a difference in the program?
The new Camera Orbit tool is very useful, also the mouse control seems to be improved over previous versions? Anyway it works well. I think it would be a good idea to add some colour sliders (like those for the light colour) on the left to change the background colour of the viewer - often models don't look best on a black background, and I'm sure a lot of people would like to be able to use a different coloured background.
As for Daz's problem, I'm gonna try exporting a really big .OBJ file and seeing if I get the same crash, if so it might be a problem with loading very dense highpoly meshes. I'll post results later.
Thanks again, good work!
Daz: OBJ's actually zip up pretty nicely, you could probably get that 75mb OBJ down to about 15mb, which isn't too bad... if you want to get that to me somehow, I could give it a test on my computer to see if it's the OBJ file that's the problem, or some sort of different, hard/software related issue?
jogshy: I tried some more stuff and I think I found another bug - the "Blur pixels" amount will go up, but not down. I can type in any value, and that works fine, but using the arrows to the right, the value only goes up and the "down" arrow doesn't have any effect.
Also I was thinking of something else, in 3dsmax there's an option to use "Supersampling" when rendering a normal map, it's like antialiasing the normal map but I guess it casts more rays and uses an average to determine the pixel colour.
Here is an example of the difference between an xNormal map (which is still very good!), and a 3dsmax "supersampled" map... the smoothness just makes it look a bit more believable.
Is it possible for you to do something like this? I'm sure it's probably not easy to do, but it'd be a great feature and I'm sure would get more people using your program for all their normal-mapping
Oh, I just thought of something else that would be very useful for the 3d viewer, too! At the moment when you hold CTRL and right-click, the camera orbits ... would it be easy to add something so that CTRL and left-click would make the light orbit? Or maybe CTRL and left-click could rotate the object itself? I think that'd be a useful feature.
Thanks again, keep up the great work!
Impressive artwork, MoP! I love your screenshots!
I think your spec/glow highlight looks different because I use a default specular power exponent of 32 and 3dsmax 16. You can control it using a specular power map in the 3D viewer ( white pure = 255 exponent = supershiny; black=0 exponent=no shiny ). Notice there is a specular map ( specular color and intensity modulation ) and a specular power map ( specular exponent ) in the 3D viewer texture browser. For more info about Phong specular exponents see these links:
http://mrl.nyu.edu/~perlin/courses/fall2005ugrad/phong.html ->See more exponent means more shiny
http://en.wikipedia.org/wiki/Specular
http://en.wikipedia.org/wiki/Phong_shading
3dsmax uses a fixed shiny exponent of 16 for the Direct3D shading viewport(i think). I use 32 by default for the full mesh unless you specify a "per-pixel-specular-power-texture" ( this is called "variable specular power", if you wanna google it, and notice is NOT the same than specular color modulation, is a different concept )
Also, I'm working to put the "HDR fullscreen glow" in the 3d previewer for the next version.
Atm, I am finishing too the "orbit light" feature ( and a chaos-animated one too ) and the change background color I think is a good suggestion, yep.
Yes, the blur thing is a bug. Ouch! Will correct it asap.
Will include the "supersampling" option as you suggest.
To Daz:
I love your 007 art!
The version 3.1.2 solved a really weird crash-critical-bug in the ambient occlusion generation. Please sure you have the 3.1.2. Ok, if you have the 3.1.2 and still crashes... Have your high poly model UVs and normals? If so, try to re-export it with ONLY positions, then use it with xNormal. Sure too you have only one mesh in your highpoly OBJ. Collapse ALL the meshes into one. Erase any materials, sure all the faces in the highpoly mesh uses only one material(the default material because you erased all the materials previously). Try to triangulate it ( perhaps is a problem with quad faces? ).
Other question... are you using xNormal in Windows and not with Wine or other Linux emulator, right?
If still crashes... Can you send me an email to [Email]jogshy [at] mixmail.com[/Email] with your operating system name and service packs, machine CPU/RAM/hardware, your xNormal XML settings, copy/paste of the exception or error from xNormal and xN_debugLog.txt log file? If you want to send me a thing with +10Mb then please use admin@santyesprogramadorynografista.net
There are no hard-coded limits. The only limit is which your machine supports. Each highpoly triangle occupies 12bytes. Each vertex 12bytes too. The normalmap width*height*20. See the xN_debugLog.txt to know the memory consumed. If you don't have enough RAM installed then expect a nice crash.. In theory Windows should enable virtual memory when physical system RAM is not enough but well... Other posibility is that xNormal goes outta "stack memory" ( 1Mb at the moment ) for the acceleration structures I use ( bsp, octrees, blah blah ). In the next version I will add a visual UI indicator to see how much memory are you using.
If you can enable me a FTP or a download link I will test your 75Mbs models and treat them confidentially. Will erase them once I solve the possible bug, I promise.
To all:
I am working too in a new install method that allows you all not to download 11Mb each time a new version comes, because is a pain, I know
And wanna thank you all for all the feedback I'm getting
This is a really awesome app btw
I am thinking about just triangulate n-verts-polygons too... Save that OBJ Daz, I will put the next version ( omg with tons of new features! ) in a few and probably will need it to test please.
Just conserve that OBJ and test it with the next version I am preparing If you want.
thx Daz.
I've been trying out the app and like the many options I am seeing!
However... I got that error when I try to load the demonhead example file, and got the same when loading one of my own objs with nmap on :
Any idea? I'm on a GF4Ti4200, XP, no service pack, 1Ghz AMD. (don't laugh!)
As for suggestions, I would love to see .ase import and sleek castshadows
Can't wait to see where that viewer is going.
"Attempted to read or write protected memory. This is often an indication that other memory is corrupt."
My computer is an athlon xp 3200 , 1 gig PC3200 ram , Ati Radion 9800 pro , running Windows XP.
Been looking for a program like this for quite awhile now, glad to see you are making one and updating it so fast
Pior you need a Radeon9xxx or GeForce FX or better to run the viewer ( or any card supporting HLSL or GLSL ). In old cards like yours the programmers must use a type of "graphics assembler" ( yes, asm, mov eax, 10 lea [eax+10] thingy! ) which is simply weird. In fact, if you see some DirectX 10 preview and the graphics cards coming, you will find that they are dropping non-HLSL shaders. Shaders are more and more complicated with the time so "assembly" paradigm is bad. We need high level shader languages (HLSL and GLSL). So I don't plan to support these old cards because I don't wanna implement a 800 lines crazy-assembly shader, sorry 8( On the other hand, I don't like the Cg approach because the Mac version I'm preparing doesn't support it and Cg+ATI=bad. GLSL is more portable I think and the way to go.
Also, to run Windows Vista(Avalon WPF) or the new Gnome 3D desktop, you will need a card with ps2.0+ or above. Even SM 3.0 won't be enough due to geometry shaders explosion ( yes we have now vertex shaders, pixel/fragment shader and geometry/topology shaders! ).
To Lupus:
Hey nice Gargantua!
What kind of mesh you use? What options you enabled? Do you have troubles running the examples or only with your own geometry? The program crashes just at start or when you try to generate a normal/displace/occlusion map? What says the xNormal_debuglog.txt?
so if you are interested in a "old hardware" path jogshy, I could support you (code it). aiming at GL 1.3, I think it would still run somewhat well on that hardware, since we are just viewing a single model... the thing that will be really tricky however is nice specular
The error happens when I try to run the program. I double click the icon and then an error window appears with that message.
If the application crashed just on start then must be a .NET 2.0 problem... Do you installed/uninstalled the Beta2 of .NET2.0? Do you use any non-latin version of WinXP? Are you using a FAT file system? Are you running Windows as administrator or as simple user?
Try to delete the xN_Settings XML files ( the files, not the folder! ). That usually is what can cause too a crash at start.
I will revise the startup code in the next version, but I am almost sure is a .NET2.0 problem... Will try to see, sorry.
I am running SP2, english version of WinXP and I think it's FAT file system (I use the default selection when I install, which I think is FAT32). I'll try deleting those files and let you know how it turns out.
No need to apologise
EDIT: Just went into the xN_Settings folder to delete what was in there, but it's already empty.
Added supersampling, light orbiting, n-sided polygons ( quads weren't enough for meh! ), UI improvements like disable buttons sounds and RAM indicator, background color slider, improved dotXSI exporter ( now supports even triangle strip lists ), no more texture coords must be in range [0,1], blur down now works and tons of bugs solved ( ms3d loading, mipmap gen, blah blah )
Ok Daz, could you test that OBJ that smashed the program some time ago now please?
Oh btw, due to the 20Gb of traffic a day I'm getting, decided to upload the program to three famous download sites.. Hope this won't be a problem.
See ya soon!
I'll download this new version and give it a try.
Thanks again, keep up the good work
edit: Ok, I just tried the new release. I have a dual monitor setup, and it stretches the viewer across both screens. It's difficult to orbit the camera this way, as my monitors aren't side by side. Can you add an option to disable it? I liked it more on my primarly monitor only.
Further to what Ely just mentioned, ability to run the viewer in a window would be nice. Currently with the viewer running, you can't Alt-Tab into Photoshop. Which means that you can only ever grab one screenshot from the viewer at a time. Unless you had a screen capture button of course ;-p
I hope you do decide to allow the viewer to run in a window, and add a button or hotkey to refresh the textures, with a checkbox to select which textures to refesh. Those two functions would make it truly a killer app for texture artists.
I've been splitting up my mesh into 3 parts and doing 3 seperate normal map passes to ensure that the rays get to every nook & cranny like so:
and then combining all 3 elements into one normal map in Photoshop. However, with this technique, surely my ambient occlusion map wont be correct will it? Since a lot of the surfaces required for occlusion aren't present for each pass. Am I making any sense?
So I conclude that for occlusion I really need to do the complete model in one go. In which case, If the rays can get to everywhere that they need to to make the occlusion map, then how come they can't for the normals, and I have to do 3 seperate passes? My head hurts!
Heres some different rays per sample I tried ( ignore the artifacting. I know what that's from ) :
So the x256 took a while but I think it was worth it. This is still invaluable to me since I can't currently use a normal map to render an AO map in maya. Perhaps theres a way, but it's really making my head hurt. I can only AO map hi poly geo.
Will add that multimonitor button and the windowed thing, no problem.
About the AO:
1)AO speed is a pain, aye. Think I have to cast 256x more rays than normal mapping. That's a lot.
The fast values are from 24 to 100. For quality 100-500. For film 1000+.
xNormal atm is 0.000000% optimized. I don't use SSE/SSE2/AMD64 instructions, use .NET ( which is like java, not pure-compiled, bytecode interpretted ) and double floating point precission... I could really speed up a bit the process ( like a 50% ), but at a TREMENDOUS memory cost ( x20 ) and some small imprecissions. I discarded that method because I could only manage a 6M poly with 1Gb of RAM, so decided to keep the slower method and optimize memory usage better than speed. If you want that just tell me, but I think is better to wait a few and consume really less memory(??? or not?). What do you prefer? Memory or speed? Let me see if I can speed up a bit the things without changing a lot.
2) I think there is no easy solution for your problem with the character, Daz
If you sepparate the model into 3 pieces the normal map could be good once you recompose it in photoshop, but I think AO won't be correct because is a global illumination technique that requires all the geometry inside the file. For normalmap, you could sepparate the model into 3 and recompose in photoshop. For ambient occlusion, you have to put all the meshes inside one file, that's sure. At the moment I'm changing a few the AO computation to speed up it a few and tolerate better the floating point goemetry.
Again, thx for all the feedback and expect other version in 2 or 3 days! ( will add the multimonitor/windowed thing, the AO map refit, the texture reload and other small UI things I have in the head atm )
Maybe the way to do both normal and occlusion with the same files would be to load the full low- and high-poly models, then allow the user to select a chunk or chunks of the lowpoly to capture normals for. They could save the resulting normal map then select the next chunk(s) to process. Choosing AO would just automatically include the full model. What do you think?
Added new ambient occlusion method ( which is better for floating geometry and 2x faster ), parallax mapping in the 3d viewer, a new example!(ugly, did myself), the multimonitor thing, ALT+tab support, backface culling, F1 to hide viewer's UI, blah blah and much more minor things.
As usually, http://www.santyesprogramadorynografista.net/projects.aspx
ps: Texture reload and ASE support for the next version, sorry.
Enjoy it!
@ Daz: there would be an easy method, but jogshy had to code it
Maya has some rendering options for object, therefore you can turn of the visibly of the object, but still let it block light...
I am thinking too to add a thing to batch-queue-render the meshes, so you don't need to recompose the nomralmap in photoshop if you use tons of floating geometry.
Well let me see what can I do.
Thanks for the update jogshy! You are indeed a machine.
So how would that work? How would you select geo? Based on seperate mesh components? Doesnt an obj export annoyingly combine all seperate objects into one model?