[ QUOTE ]
[ QUOTE ]
... the height map generator seems to not work so well. All it gives me is a flat black and white image, more like a mask than a height map. And there doesn't seem to be any options i can change here.
[/ QUOTE ]
Probably you need to set the min / max height in the heightmap's options. To know this distance you can use a 3dsmax's tape or set it to aprox the radius of the object.
[ QUOTE ]
XSIFtk_6_0.dll not found
[/ QUOTE ]
Ouchies... gonna patch it asap, sorry.
[/ QUOTE ]
Theres settings for min/max value, why wouldnt you simply use the cage offset value for min/max height? This feature worked perfectly fine without any need for extra settings in older versions.
[edit] That did fix it, its a little confusing tho, its labled min/max value, one would assume that is just for clamping the values, not a setting related to height.
This feature worked perfectly fine without any need for extra settings in older versions.
[/ QUOTE ]
Well, previous version uses a thing called "auto-normalization". Basically were exploring the max height and then divide all values using that value... The problem was that some points could be too far and others too close... so the result could be too black/white and will loose accuracy.
[ QUOTE ]
Theres settings for min/max value, why wouldnt you simply use the cage offset value for min/max height?
[/ QUOTE ]
Hmmm... you're right. Perhaps I could use the cage or front/back ray distances if the cage is not available. But i'm not sure how to "normalize" the results... Probably would need to get the maximum cage's distance... and then divide all the heights by it? Well, gonna test that for the 3.14.1 Beta 1 soon.
I just tested out the fast ao, save to .sbm, with a silo .sib. I have a sculpted mesh with about 4 levels but Xnormal saves the lowest (level 0) instead of the highest to .sbm.
[ QUOTE ]
I just tested out the fast ao, save to .sbm, with a silo .sib. I have a sculpted mesh with about 4 levels but Xnormal saves the lowest (level 0) instead of the highest to .sbm.
[/ QUOTE ]
I suspect the Silo mesh format only stores the base mesh as data I can read. Subdivision data and algorithm is still a mystery.
Btw, got some progress eyecandy:
The starship with HDR:
Sethodie's car for Gameartisan Apocalyse competition, overexposed:
Smiley... under-exposed:
Smiley... ok-exposed:
Teapot HDR + St.PetersCubemap.hdr:
Just need to finish shadows to release the 3.14.1 Beta 1.
Btw, i'm looking for some massive vegetation example to test a new MSAA thingy if somebody is interested.
Been using 3.14 and i'd just like to say that the memory improvements and speed improvements are quite amazing. This saves me sooooo much time and headaches day to day. Thanks so much santiago!!! Never do i have to worry about running out of memory, or wait more than a couple of minutes for my maps to render. The realtime preview is a huge time save as well, i can see instantly if errors are going to show up and cancle the bake.
[ QUOTE ]
Jogshy: very nice. mind passing on a few tips on how you're handling shadows?
[/ QUOTE ]
Well, originally was using stencil shadows with wedges... but used too much fillrate. Then I tryed a GPU raytracer like the one i'm using the simple GPU AO tool... but got only 5FPS. Also tryed precomputed PRT... but required to tessellate a lot the scene and was very slow to compute.
Then I decided to use a shadowmap with simple penumbra estimation. I don't have many options because wanna use a point omni light. I could implement some kind of parallel split algorigthm but they are good only for outdoor environments 8(
On the other hand, true penumbra shadows with PCSS are too slow for the average graphics cards. I tryed but framerate dropped severely. Needed to use 64 samples per pixel to get a decent result... and was terribly slow even with a 8800GT.
Precision and bias were a problem. Some graphics cards don't allow FP32 textures/blending/filtering. I tryed a second-depth approach but didn't solve the biasing problems completely. Finally, used a variation of the Markov-Chevyschev upper bound equation to mitigate it... but adds terrible "light bleeding"... So better i'm letting the artist to tweak it with a slider...
For some reason when you talk like that, Santy, I get slightly aroused [in a completely heterosexual way, of course]. I don't understand anything what you're saying, but damn it's sexy [in a completely heterosexual way, of course].
Anyways, things just keep getting better and better. One question, though, and pardon my ignorance here.. DX10 shadows? I'm guessing we're going to have to do the whole Vista thing then..?
[ QUOTE ]
One question, though, and pardon my ignorance here.. DX10 shadows? I'm guessing we're going to have to do the whole Vista thing then..?
[/ QUOTE ]
Hehe, I'm coding a DX10 graphics driver and DX10 is Vista-only, yep. I plan to code an OpenGL 3.1 Mount Evans(which should work in W98/W2k/XP) with the same(or even better) effects than the DX10 one... but needed to do first the DX10 to learn how to use the new hardware features
Most of those effects could be ported to DX9 too... but will be a little complicated and slow.
That's a bit disappointing to hear, but -- and this is weird -- I feel compelled to actually install Vista (willingly) on my box just for the xNormal viewer with DX10 shadows, until you get the OGL 3.1 stuff sorted. Not even Crysis made me feel compelled to install Vista.
There you have it. xNormal owns Crysis. Confirmed.
Great job, as usual Oh, one thing I've been wondering about. What motivates you to keep developing xNormal?
- Added the DX10 graphics driver with improved HDRI, glow effect, logarithmic tone mapping and MSAA transparency support. Unfortunally, as you should know, DX10 is Vista-only. However, I plan an OpenGL 3.1 Mount Evans driver once the spec is released.
- The height map rendering no longer requires a numerical value to set the min/max search distance. Now it uses the cages or the uniform ray distance values.
- Added an useul text inside the 3D viewer to see the maximum cage extrussion.
- Solved some bugs, including a crash in the Simple GPU AO tool when you try to load a file with multiple meshes embedded.
Btw, i'm looking for a new example with some kind of vegetation/HDR skydome or reflection in order to test the new DX10 things. Please, send me a PM or an e-mail if interested.
Hey Im using xnormal for the first time today and Im finding it a bit difficult. I load my high poly and low poly then I adjust the cage and save the mesh.
I got a high poly sbm and a low poly sbm. I load these as high and low poly meshes and in the 3D view it all looks good and theres nothing coming through the cage but when I hit bake I get massive holes in the normal map??
use cage is definitely already on. Heres what I get in the viewer, can anyone help? I dont see anything wrong but maybe Im just not using xnormal properly.
[ QUOTE ]
Hey Im using xnormal for the first time today and Im finding it a bit difficult.
[/ QUOTE ]
Well, xNormal has many features and can be a little confusing at the start. Take a quick look ( just a quick one ) at the documentation, tutorials and examples. Once you do that just try to bake your first normal map. A typical workflow is:
1. Fill one slot of the highpoly list ( million polys mesh ) with, for example, myHighpoly.obj
2. Fill one slot of the lowpoly list ( realtime model ), with for example myLowpoly.obj. This needs to be UV-mapped with non-overlapped UVs(aka = a good UV unwrapped model ).
3. Enter the 3D viewer. Check the "show cage" option. Extrude it until covers completely the highpoly model. Press the "Save meshes" button ( that will generate two .SBM files.. one with the highpoly that you can ignore and other with the lowpoly + cage info which is the important ).
4. Clean your mesh list ( the myHighpoly.obj and the myLowpoly.obj ) and assign the SBMs instead ( which contain cage info ). Sure you check the "use cage" option or the cage information present in the SBM won't be used ( will use uniform ray distances if you uncheck the "use cage" option ).
5. Set your basic normal map options ( output image size, tangent space option, output file name, map type to generate ( a normal map in this case ), etc.. )
6. Generate the map.
You can avoid some steps if you use 3dsmax because you can edit the cage there and export it to xNormal using the SBM plugin... or model a cage and save it as myCage.obj and use it as "external cage"... or you can ignore cages method and set the uniform ray distance method ( but only works well for very simple meshes )... but cages are the best method.
Once you know how to do all this you can play a few with the xNormal tools ( height map to normal map converter, simple AO baker, Photoshop plugins, etc... )
[ QUOTE ]
I load my high poly and low poly then I adjust the cage and save the mesh.
I got a high poly sbm and a low poly sbm. I load these as high and low poly meshes and in the 3D view it all looks good and theres nothing coming through the cage but when I hit bake I get massive holes in the normal map??
[/ QUOTE ]
Yep, sure you set the "use cage" option in the lowpoly mesh list... or the cage data will be ignored when the .SBM mesh file loads. By default xNormal uses the "uniform ray distances" method and not the cages one.
Also, caution with the "discard back-faces hits" option... That discards hits if the highpoly and lowpoly normals are opposite ( too perpendicular ).
Other interesting option if the "Smooth normals" in the lowpoly mesh list slots... Basically it averages(and unifyies) the vertex normals automatically... useful for some patch/subdiv models because some programs flips the normals or face order.
And, finally, a very important advice.... Sure you do a "freeze transformations" in Maya or ResetXForm in Max BEFORE to export.... and... the max2obj OBJ exporter that comes with Max is baaaaaaaad(messes the vertex normals)... better export as ASE, 3DS or SBM if you work with Max.
If you can, use always .SBM files which are the native binary xNormal file format. Loads faster than any other and contains all the mesh data that xNormal needs ( vertices, normals with smoothing groups support, tangent basis, cage info, UVs, occlusion, etc ).
The OVB was the old xNormal format... in XML-human readable text... so can be slow to load and is not prepared for dense meshes.
Ofc, you can use OBJ, 3DS, ASE and other formats too... but will load slower and some cannot store smoothing groups or normals and neither cage info or per-vertex occlusion.
That's all you need to know to start. Hope it helps.
Hey thanks for the reply Jogshy. I have found some reversed normals in my mesh and fixed them and tried again but the result is almost the same. When I turn "discard back-faces hits" off then the areas which are blank then have bright green normal mapping stuff in them which looks bad?. Any idea whats going on? is it my lowpoly slightly going through my high poly in places because i built it that way? As in the screenshot before nothing is going through the cage.
[ QUOTE ]
I have found some reversed normals in my mesh and fixed them and tried again but the result is almost the same. When I turn "discard back-faces hits" off then the areas which are blank then have bright green normal mapping stuff in them which looks bad?. Any idea whats going on?
[/ QUOTE ]
Some programs have the bad habit to flip patch/subdiv normals ( Silo more specifically ).
Your cage seems to be ok... so try to set the "Smooth normals" in the lowpoly mesh slot options.. That should average and unify the lowpoly mesh normals.
The "Ignore back face hits" thing usually keep it enabled always to avoid problems.
I see your model's radius is a bit small(1.2)... Radius aprox to 1.0 is conflictive... Better scale the model up using the "Mesh scale" option in Xnormal ( which avoids the need to re-export the model ). Usually is better to have meshes with a radius >10 to avoid numerical stability problems )
Btw... what program you used to export the meshes?
OH WOW! that was it! its the SCALE! So the mesh needs to be bigger in order for xnormal to do its stuff cant believe Ive spent nearly all afternoon on this but now I know how to use xnormal for normal maps on anything lol I downloaded and followed alot of those tutorials! thanks alot Jogshy.
Oh I used cinema 4d with the riptide plugin for exporting good objs. The high poly is from zbrush
[/ QUOTE ]
Told ya! Usually xNormal shows a warning indicating this if radius <1.0f... you had bad luck... was 1.2 so didn't see the message box hehe... I should increase the warning to 2.0f it seems.
The cause is that I use an "epsilon" error margin for floating point calculations of 0.01f to cut planes, point in box error, etc... also use a IEEE data model known as "half precision" which don't like small numbers and neither +-2048.0f. Usually a radius in range [10,2048] is fine... is not, just use the "scale mesh" option so you avoid to re-export the models... had auto-scale some time ago... but I had to eliminate it to maintain the proportions the artist wanted for some operations.
[ QUOTE ]
The high poly is from zbrush
[/ QUOTE ]
Remember to set the eFlipX and eFlipY in the Zbrush's OBJ export options. In that way the mesh will be in the OpenGL coordinate system ( and ccw vertex order = front face ) that xNormal(and Maya and XSI) uses.
Btw, you can generate a normal map from a ZBrushed highpoly easy ( but less accurate than the lowpoly-highpoly method ) from a displacement map like is said here:
Hi, I'm having some problems with Xnormal lately. It looks like there's no normal map on the models (even examples).
I think it's because my graphics card, which should be compatible with SM3, but it has problems sometimes.
Is there some way how to force Xnormal to use SM2?
Thanks
[ QUOTE ]
Hi, I'm having some problems with Xnormal lately. It looks like there's no normal map on the models (even examples).
I think it's because my graphics card, which should be compatible with SM3, but it has problems sometimes.
Is there some way how to force Xnormal to use SM2?
Thanks
[/ QUOTE ]
It works with SM4, 3, 2 and 1 but the graphics card need enough VRAM(32Mb as absolute minimum).
If you choose the OpenGL graphics driver in the plugin manager, need a SM2 GLSL-compatible card with OpenGL 1.3 support ( GFFX, Rad9 or above ).
If you choose the DX9 graphics driver, will use automatically SM1 for GF3/4/Rad8, SM2 for GFFX/Rad9, or SM3 for GF6/7/RadX1 or above cards.
If you have Vista, you can choose the DX10 graphics driver which will use SM4.
Try always to use updated drivers(Forceware/Catalyst), specially if you work with the OpenGL graphics driver or you use a laptop.
If you see the model without normal map, but is assigned, sure you didn't set the "Matte" option ( because annuls the dynamic lighting ). Also sure you set well the "tangent space" option ( to indicate the viewer if the normal map is in object space or tangent space ).
If you post an screenshot of the results + viewer settings I could tell you more.
Thanks a lot for answer.
At the end problem wasn't in the shader model, but in size of textures . I have Radeon x1650 -256MB. When I use 3 1024x1024 .jpegs it's alright but 3 .tgas are problems. (Strange, I thougth textures are sent to graphic card in uncompressed format)
[ QUOTE ]
Thanks a lot for answer.
At the end problem wasn't in the shader model, but in size of textures . I have Radeon x1650 -256MB. When I use 3 1024x1024 .jpegs it's alright but 3 .tgas are problems. (Strange, I thougth textures are sent to graphic card in uncompressed format)
[/ QUOTE ]
See if the "use texture compression" is enabled. Better enable it if your graphics card does not have much VRAM(you can see the used VRAM in the 3d viewer... a text near the right-bottom part of the screen ).
Do the TGA have alpha channels btw? Before the 3.14.0 the alpha channel in the normal map could contain a height map used for parallax mapping. After, I modified it to allow the artist to use a separated height map file... perhaps is a bug I need to get rid of... but if you have the "use parallax mapping" disabled should work ok in theory even if this is a bug.
Hey I deserve to be slapped in my face. I overlooked, the tgas was 2048x2048. With 1024 it works. But still, 3x2048x2048 takes 36 MB on hard drive and about 76MB VRAM - and my 256MB RAM card is not able to handle it
NM displays only when VRAM usage is around 32MB or below.
Same with/without alpha channels.
Ok, after alot of people saying that I should check out Xnormal I am checking it out. Although alot of the technical talk within this thread is leaving me in the dust:) I followed the Zbrush to Xnormal tutorial on your site Jogshy, but was curious why he did a displacement map, to turn it into a normal, instead of having a normal from zbrush and xnormal checking it over. Also, how high of a mesh can Xnormal cast from, as with max, I cannot bring anything in over 2 million to cast from before my computer crashes, and I lost my polycruncher a long time ago. Just curious, and thanks in advance for the help.
Spark: should be able to handle VERY high res meshs, i think i'e personally tried 5 million poly objs, Santiago was testing it out with models in upwards of 100 million tris tho, so i would just say export your highest res mesh you have and try it out
I easily rendered a 12 millions tris model with xnormal, and that was before all the memory optimizations and gpu AO, so I guess that's pretty safe now
@tom.drin: I suspect it can be a bug. Gonna send you a modified lite version soon. Try to disable the AA and change from/to the DX9 graphics driver to the OGL one while, pls.
@Spark: With around 2Gb(free, not total) I was able to create a normal map for the Lucy mesh (114M points, 34M poly mesh) sucessfully. It depends on geometry though.
On the other hand, the Simple AO GPU tool usually cannot manage meshes greater than 1-4M polys... it's a HW limitation.
Wow, that is some pretty intense meshes you guys are dealing with, so yah I think I won't be hitting that high so I should be good:) Thanks for the responses guys, and if there are any more examples for tutorials I would love to check them out.
Released xNormal 3.14.2. Just corrected some minor bugs, the DX9 graphics driver consumes less VRAM, added the highpoly "smooth normals" option and "mesh scale" in the SImple AO GPU tool. Nothing really important.
[ QUOTE ]
Hey, question... Where's the problem when it gives this error: 'Unexpected error rendering the triangles for maps'
And how to fix it?
[/ QUOTE ]
Hard to say. Need more info:
When and where happens that? xNormal version? OS and HW config? Screenshots? What says the xNormal_debug.txt on MyDocuments\xNormal? Can you view and render well the examples?
[/ QUOTE ]
Huh... you are using an unstable beta version... and really old(years old one)! So is normal you are getting problems . I recommend you to use the latest non-beta one... like the 3.14.2.
I'll post this here so it dosent get lost forever in my other thread, refering to the "fine detail" panel.
Is that what that panel is for? Adding a bumpmap for the lowpoly's uvs? Or is it for the highpolys? I've never really understood what that was for, infact i thought it was a feature you abandoned a while ago.
In any case it would be great if there was an option somewhere to combine a bumpmap/normals into an object space map, either while rendering or afterwords!
[ QUOTE ]
I'll post this here so it dosent get lost forever in my other thread, refering to the "fine detail" panel.
Is that what that panel is for? Adding a bumpmap for the lowpoly's uvs? Or is it for the highpolys? I've never really understood what that was for, infact i thought it was a feature you abandoned a while ago.
[/ QUOTE ]
Is to add fine detail to the output normal map. You set there a heightmap so will be converted to a tangent space map. Then, will be combined with the output normal map from the rendering process (performing automatic tangent/object space conversion). The "blend factor" controls the mix amount. The "scale factor" is used only for height maps and ignored for normal maps.
That can add fine detail to object space maps or tangent space maps too. See how is set in the Smiley example.
But there is a problem with that... currently you cannot control really the "strength" of the heightmap and neither the heightmap-to-normal map conversion(will use always a 4 neighbors method ). I'm gonna try to improve that for the next release.
Jogshy, would it be possible to add wacom tablet support for the 3d viewer? The cursor tends to get lost when navigating in tablet mode (works fine in mouse mode though).
[ QUOTE ]
Jogshy, would it be possible to add wacom tablet support for the 3d viewer? The cursor tends to get lost when navigating in tablet mode (works fine in mouse mode though).
[/ QUOTE ]
Well, I don't have a Wacom tablet so would be hard to test. If I get one some day I will add support for it, that's for sure.
Replies
[ QUOTE ]
... the height map generator seems to not work so well. All it gives me is a flat black and white image, more like a mask than a height map. And there doesn't seem to be any options i can change here.
[/ QUOTE ]
Probably you need to set the min / max height in the heightmap's options. To know this distance you can use a 3dsmax's tape or set it to aprox the radius of the object.
[ QUOTE ]
XSIFtk_6_0.dll not found
[/ QUOTE ]
Ouchies... gonna patch it asap, sorry.
[/ QUOTE ]
Theres settings for min/max value, why wouldnt you simply use the cage offset value for min/max height? This feature worked perfectly fine without any need for extra settings in older versions.
[edit] That did fix it, its a little confusing tho, its labled min/max value, one would assume that is just for clamping the values, not a setting related to height.
This feature worked perfectly fine without any need for extra settings in older versions.
[/ QUOTE ]
Well, previous version uses a thing called "auto-normalization". Basically were exploring the max height and then divide all values using that value... The problem was that some points could be too far and others too close... so the result could be too black/white and will loose accuracy.
[ QUOTE ]
Theres settings for min/max value, why wouldnt you simply use the cage offset value for min/max height?
[/ QUOTE ]
Hmmm... you're right. Perhaps I could use the cage or front/back ray distances if the cage is not available. But i'm not sure how to "normalize" the results... Probably would need to get the maximum cage's distance... and then divide all the heights by it? Well, gonna test that for the 3.14.1 Beta 1 soon.
1. Automatic height min/max distance using the cages or uniform ray distances ( done )
2. An indicator inside the 3D viewer to show the maximum cage extrussion ( done )
3. Solve bugs from 3.14.0 ( done? )
4. The Dx10 graphics driver with improved effects ( almost done
I just tested out the fast ao, save to .sbm, with a silo .sib. I have a sculpted mesh with about 4 levels but Xnormal saves the lowest (level 0) instead of the highest to .sbm.
[/ QUOTE ]
I suspect the Silo mesh format only stores the base mesh as data I can read. Subdivision data and algorithm is still a mystery.
Btw, got some progress eyecandy:
The starship with HDR:
Sethodie's car for Gameartisan Apocalyse competition, overexposed:
Smiley... under-exposed:
Smiley... ok-exposed:
Teapot HDR + St.PetersCubemap.hdr:
Just need to finish shadows to release the 3.14.1 Beta 1.
Btw, i'm looking for some massive vegetation example to test a new MSAA thingy if somebody is interested.
(excuse the E.T's effect on finger... I was playing with the emissive channel
Gonna release the 3.14.1 b1 soon.
Jogshy: very nice. mind passing on a few tips on how you're handling shadows?
[/ QUOTE ]
Well, originally was using stencil shadows with wedges... but used too much fillrate. Then I tryed a GPU raytracer like the one i'm using the simple GPU AO tool... but got only 5FPS. Also tryed precomputed PRT... but required to tessellate a lot the scene and was very slow to compute.
Then I decided to use a shadowmap with simple penumbra estimation. I don't have many options because wanna use a point omni light. I could implement some kind of parallel split algorigthm but they are good only for outdoor environments 8(
On the other hand, true penumbra shadows with PCSS are too slow for the average graphics cards. I tryed but framerate dropped severely. Needed to use 64 samples per pixel to get a decent result... and was terribly slow even with a 8800GT.
Precision and bias were a problem. Some graphics cards don't allow FP32 textures/blending/filtering. I tryed a second-depth approach but didn't solve the biasing problems completely. Finally, used a variation of the Markov-Chevyschev upper bound equation to mitigate it... but adds terrible "light bleeding"... So better i'm letting the artist to tweak it with a slider...
Anyways, things just keep getting better and better. One question, though, and pardon my ignorance here.. DX10 shadows? I'm guessing we're going to have to do the whole Vista thing then..?
One question, though, and pardon my ignorance here.. DX10 shadows? I'm guessing we're going to have to do the whole Vista thing then..?
[/ QUOTE ]
Hehe, I'm coding a DX10 graphics driver and DX10 is Vista-only, yep. I plan to code an OpenGL 3.1 Mount Evans(which should work in W98/W2k/XP) with the same(or even better) effects than the DX10 one... but needed to do first the DX10 to learn how to use the new hardware features
Most of those effects could be ported to DX9 too... but will be a little complicated and slow.
There you have it. xNormal owns Crysis. Confirmed.
Great job, as usual
The HDR causes a caustics effect with the translucent shadows on the wings's projection hahah!
I'm not sure if this is good or bad hahaha! Well, I bet you could control it with a slider i'm gonna add for the bright HDR pass threshold.
- Added the DX10 graphics driver with improved HDRI, glow effect, logarithmic tone mapping and MSAA transparency support. Unfortunally, as you should know, DX10 is Vista-only. However, I plan an OpenGL 3.1 Mount Evans driver once the spec is released.
- The height map rendering no longer requires a numerical value to set the min/max search distance. Now it uses the cages or the uniform ray distance values.
- Added an useul text inside the 3D viewer to see the maximum cage extrussion.
- Solved some bugs, including a crash in the Simple GPU AO tool when you try to load a file with multiple meshes embedded.
Get it, as usually, from the downloads section of http://www.xnormal.net
Btw, i'm looking for a new example with some kind of vegetation/HDR skydome or reflection in order to test the new DX10 things. Please, send me a PM or an e-mail if interested.
thx!
Thanks
I got a high poly sbm and a low poly sbm. I load these as high and low poly meshes and in the 3D view it all looks good and theres nothing coming through the cage but when I hit bake I get massive holes in the normal map??
Hey Im using xnormal for the first time today and Im finding it a bit difficult.
[/ QUOTE ]
Well, xNormal has many features and can be a little confusing at the start. Take a quick look ( just a quick one ) at the documentation, tutorials and examples. Once you do that just try to bake your first normal map. A typical workflow is:
1. Fill one slot of the highpoly list ( million polys mesh ) with, for example, myHighpoly.obj
2. Fill one slot of the lowpoly list ( realtime model ), with for example myLowpoly.obj. This needs to be UV-mapped with non-overlapped UVs(aka = a good UV unwrapped model ).
3. Enter the 3D viewer. Check the "show cage" option. Extrude it until covers completely the highpoly model. Press the "Save meshes" button ( that will generate two .SBM files.. one with the highpoly that you can ignore and other with the lowpoly + cage info which is the important ).
4. Clean your mesh list ( the myHighpoly.obj and the myLowpoly.obj ) and assign the SBMs instead ( which contain cage info ). Sure you check the "use cage" option or the cage information present in the SBM won't be used ( will use uniform ray distances if you uncheck the "use cage" option ).
5. Set your basic normal map options ( output image size, tangent space option, output file name, map type to generate ( a normal map in this case ), etc.. )
6. Generate the map.
You can avoid some steps if you use 3dsmax because you can edit the cage there and export it to xNormal using the SBM plugin... or model a cage and save it as myCage.obj and use it as "external cage"... or you can ignore cages method and set the uniform ray distance method ( but only works well for very simple meshes )... but cages are the best method.
Once you know how to do all this you can play a few with the xNormal tools ( height map to normal map converter, simple AO baker, Photoshop plugins, etc... )
[ QUOTE ]
I load my high poly and low poly then I adjust the cage and save the mesh.
I got a high poly sbm and a low poly sbm. I load these as high and low poly meshes and in the 3D view it all looks good and theres nothing coming through the cage but when I hit bake I get massive holes in the normal map??
[/ QUOTE ]
Yep, sure you set the "use cage" option in the lowpoly mesh list... or the cage data will be ignored when the .SBM mesh file loads. By default xNormal uses the "uniform ray distances" method and not the cages one.
Also, caution with the "discard back-faces hits" option... That discards hits if the highpoly and lowpoly normals are opposite ( too perpendicular ).
Other interesting option if the "Smooth normals" in the lowpoly mesh list slots... Basically it averages(and unifyies) the vertex normals automatically... useful for some patch/subdiv models because some programs flips the normals or face order.
And, finally, a very important advice.... Sure you do a "freeze transformations" in Maya or ResetXForm in Max BEFORE to export.... and... the max2obj OBJ exporter that comes with Max is baaaaaaaad(messes the vertex normals)... better export as ASE, 3DS or SBM if you work with Max.
If you can, use always .SBM files which are the native binary xNormal file format. Loads faster than any other and contains all the mesh data that xNormal needs ( vertices, normals with smoothing groups support, tangent basis, cage info, UVs, occlusion, etc ).
The OVB was the old xNormal format... in XML-human readable text... so can be slow to load and is not prepared for dense meshes.
Ofc, you can use OBJ, 3DS, ASE and other formats too... but will load slower and some cannot store smoothing groups or normals and neither cage info or per-vertex occlusion.
That's all you need to know to start. Hope it helps.
I have found some reversed normals in my mesh and fixed them and tried again but the result is almost the same. When I turn "discard back-faces hits" off then the areas which are blank then have bright green normal mapping stuff in them which looks bad?. Any idea whats going on?
[/ QUOTE ]
Some programs have the bad habit to flip patch/subdiv normals ( Silo more specifically ).
Your cage seems to be ok... so try to set the "Smooth normals" in the lowpoly mesh slot options.. That should average and unify the lowpoly mesh normals.
The "Ignore back face hits" thing usually keep it enabled always to avoid problems.
I see your model's radius is a bit small(1.2)... Radius aprox to 1.0 is conflictive... Better scale the model up using the "Mesh scale" option in Xnormal ( which avoids the need to re-export the model ). Usually is better to have meshes with a radius >10 to avoid numerical stability problems )
Btw... what program you used to export the meshes?
Oh I used cinema 4d with the riptide plugin for exporting good objs. The high poly is from zbrush
OH WOW! that was it! its the SCALE!
[/ QUOTE ]
Told ya! Usually xNormal shows a warning indicating this if radius <1.0f... you had bad luck... was 1.2 so didn't see the message box hehe... I should increase the warning to 2.0f it seems.
The cause is that I use an "epsilon" error margin for floating point calculations of 0.01f to cut planes, point in box error, etc... also use a IEEE data model known as "half precision" which don't like small numbers and neither +-2048.0f. Usually a radius in range [10,2048] is fine... is not, just use the "scale mesh" option so you avoid to re-export the models... had auto-scale some time ago... but I had to eliminate it to maintain the proportions the artist wanted for some operations.
[ QUOTE ]
The high poly is from zbrush
[/ QUOTE ]
Remember to set the eFlipX and eFlipY in the Zbrush's OBJ export options. In that way the mesh will be in the OpenGL coordinate system ( and ccw vertex order = front face ) that xNormal(and Maya and XSI) uses.
Btw, you can generate a normal map from a ZBrushed highpoly easy ( but less accurate than the lowpoly-highpoly method ) from a displacement map like is said here:
http://3dsmaxer.blogspot.com/2007/06/exporting-high-quality-nomal-maps-in.html
As for vegetation you could try this out :
http://graphics.uni-konstanz.de/~luft/ivy_generator/
Good luck!
OH WOW! that was it! its the SCALE!
[/ QUOTE ]
Holy crap. That solved -my- problem as well, that I've been having. I need to learn to ask for help. Awesome, thanks, both of you
Great about the new release as well. Going to install Vista tomorrow (still feels weird writing that..) and give it a spin.
I think it's because my graphics card, which should be compatible with SM3, but it has problems sometimes.
Is there some way how to force Xnormal to use SM2?
Thanks
Hi, I'm having some problems with Xnormal lately. It looks like there's no normal map on the models (even examples).
I think it's because my graphics card, which should be compatible with SM3, but it has problems sometimes.
Is there some way how to force Xnormal to use SM2?
Thanks
[/ QUOTE ]
It works with SM4, 3, 2 and 1 but the graphics card need enough VRAM(32Mb as absolute minimum).
If you choose the OpenGL graphics driver in the plugin manager, need a SM2 GLSL-compatible card with OpenGL 1.3 support ( GFFX, Rad9 or above ).
If you choose the DX9 graphics driver, will use automatically SM1 for GF3/4/Rad8, SM2 for GFFX/Rad9, or SM3 for GF6/7/RadX1 or above cards.
If you have Vista, you can choose the DX10 graphics driver which will use SM4.
Try always to use updated drivers(Forceware/Catalyst), specially if you work with the OpenGL graphics driver or you use a laptop.
If you see the model without normal map, but is assigned, sure you didn't set the "Matte" option ( because annuls the dynamic lighting ). Also sure you set well the "tangent space" option ( to indicate the viewer if the normal map is in object space or tangent space ).
If you post an screenshot of the results + viewer settings I could tell you more.
Hope it helps.
At the end problem wasn't in the shader model, but in size of textures
Thanks for help and this superb software again.
Thanks a lot for answer.
At the end problem wasn't in the shader model, but in size of textures
[/ QUOTE ]
See if the "use texture compression" is enabled. Better enable it if your graphics card does not have much VRAM(you can see the used VRAM in the 3d viewer... a text near the right-bottom part of the screen ).
Do the TGA have alpha channels btw? Before the 3.14.0 the alpha channel in the normal map could contain a height map used for parallax mapping. After, I modified it to allow the artist to use a separated height map file... perhaps is a bug I need to get rid of... but if you have the "use parallax mapping" disabled should work ok in theory even if this is a bug.
NM displays only when VRAM usage is around 32MB or below.
Same with/without alpha channels.
Spark
@Spark: With around 2Gb(free, not total) I was able to create a normal map for the Lucy mesh (114M points, 34M poly mesh) sucessfully. It depends on geometry though.
On the other hand, the Simple AO GPU tool usually cannot manage meshes greater than 1-4M polys... it's a HW limitation.
Spark
And how to fix it?
Hey, question... Where's the problem when it gives this error: 'Unexpected error rendering the triangles for maps'
And how to fix it?
[/ QUOTE ]
Hard to say. Need more info:
When and where happens that? xNormal version? OS and HW config? Screenshots? What says the xNormal_debug.txt on MyDocuments\xNormal? Can you view and render well the examples?
xnormal version:
3.10.4.20398 Beta 1
graphicscard:
http://bbox.sylorn.com/random/gfxc.jpg
rest of the info you need:
http://bbox.sylorn.com/random/cpuz.htm
xnormal_debug.txt:
Sun Feb 11 00:58:02 2007
Trying to allocate 0.264 Mb in virtual memory for highpoly vertices/normals/UVs...
Sun Feb 11 00:58:02 2007
Trying to allocate 0.880 Mb in virtual memory for highpoly triangles...
xnormal version:
3.10.4.20398 Beta 1
[/ QUOTE ]
Huh... you are using an unstable beta version... and really old(years old one)! So is normal you are getting problems
[ QUOTE ]
Sun Feb 11 00:58:02 2007
[/ QUOTE ]
I think you mean 2008
About 2007... Yeah it seems I put my computers time back to 2007 for some reason...
Is that what that panel is for? Adding a bumpmap for the lowpoly's uvs? Or is it for the highpolys? I've never really understood what that was for, infact i thought it was a feature you abandoned a while ago.
In any case it would be great if there was an option somewhere to combine a bumpmap/normals into an object space map, either while rendering or afterwords!
I'll post this here so it dosent get lost forever in my other thread, refering to the "fine detail" panel.
Is that what that panel is for? Adding a bumpmap for the lowpoly's uvs? Or is it for the highpolys? I've never really understood what that was for, infact i thought it was a feature you abandoned a while ago.
[/ QUOTE ]
Is to add fine detail to the output normal map. You set there a heightmap so will be converted to a tangent space map. Then, will be combined with the output normal map from the rendering process (performing automatic tangent/object space conversion). The "blend factor" controls the mix amount. The "scale factor" is used only for height maps and ignored for normal maps.
That can add fine detail to object space maps or tangent space maps too. See how is set in the Smiley example.
But there is a problem with that... currently you cannot control really the "strength" of the heightmap and neither the heightmap-to-normal map conversion(will use always a 4 neighbors method ). I'm gonna try to improve that for the next release.
Jogshy, would it be possible to add wacom tablet support for the 3d viewer? The cursor tends to get lost when navigating in tablet mode (works fine in mouse mode though).
[/ QUOTE ]
Well, I don't have a Wacom tablet so would be hard to test. If I get one some day I will add support for it, that's for sure.