Jason Posted December 21, 2005 Author Share Posted December 21, 2005 Pretty cool but I suspect that is without Raytracing? I can get the same results with Mental Ray with Ray Tracing in about the same time. The thing is that Mental Ray chokes on Displacements where as Mantra does not. So there is always a trade it seems. If Mantra could become a faster Ray Tracer then that would certainly help when it comes to what renderer to use. Cheers, Nate Nesler 23363[/snapback] No, it is raytracing, 32 times per sample. There is no way to calculate ambient occlusion in Mantra without raytracing (without developing your own methods like the one Mario pointed out). Raytracing in Mantra is very fast - comparable to Mental Ray. Also. when I say "per sample", this means that it'll fire a raytracing sample per (unoccluded) micropoly vertex - i.e., if there are 20 poly vertices crowded into a single pixel (and in this case, very definately considering the resolution of the model) then it'll still fire at least 20x32 rays. In actuality it may be firing even more rays because it super-samples the raytracing calls by firing rays from the micropoly interiors too. If I set mantra into pure raytracing mode then it may fire fewer rays overall in this case. I could make a test of that too., Quote Link to comment Share on other sites More sharing options...
Jason Posted December 21, 2005 Author Share Posted December 21, 2005 And for those with morbid curiosity, here are a couple of screen grabs of the radio dials and the back of the ignition unit. Quote Link to comment Share on other sites More sharing options...
MatrixNAN Posted December 21, 2005 Share Posted December 21, 2005 No, it is raytracing, 32 times per sample. There is no way to calculate ambient occlusion in Mantra without raytracing (without developing your own methods like the one Mario pointed out). Raytracing in Mantra is very fast - comparable to Mental Ray.Also. when I say "per sample", this means that it'll fire a raytracing sample per (unoccluded) micropoly vertex - i.e., if there are 20 poly vertices crowded into a single pixel (and in this case, very definately considering the resolution of the model) then it'll still fire at least 20x32 rays. In actuality it may be firing even more rays because it super-samples the raytracing calls by firing rays from the micropoly interiors too. If I set mantra into pure raytracing mode then it may fire fewer rays overall in this case. I could make a test of that too., 23364[/snapback] Wow that is totally Awesome. I would still like some more nodes for better control over reflections, and specular values in VOPS. I have to admit that I am still not super strong in Mantra Renderman even though I have learned a lot. I am stronger in XSI's Mental Ray Shading Trees. I have a way easier time manipulating the light to surface interaction in their trees because of a few very useful nodes for manipulating Ray Tracing. However, they still don't have the high degree of control over the specular the way they do over the reflections. I could post a more formal layout and definition of nodes if anyone would be interested. There is also a desire for Fuzzy Logic, Bayesian Learning, and Neural Nets VOPS ok now I go run and hide. Would be really useful for POPS, SOPS, and VOPS Shading. lol Cheers, Nate Nesler Quote Link to comment Share on other sites More sharing options...
MatrixNAN Posted December 21, 2005 Share Posted December 21, 2005 And for those with morbid curiosity, here are a couple of screen grabs of the radio dials and the back of the ignition unit. 23365[/snapback] Its really detailed, but the modeling Topology is not clean. Its a car manufacture so they probably don't care. lol Did they do a laser scan of the car parts and then do a poly reduction on them or did someone model this car??? Very cool man. Cheers, Nate Nesler Quote Link to comment Share on other sites More sharing options...
pbowmar Posted December 21, 2005 Share Posted December 21, 2005 Please post the layout from XSI if you don't mind. VOPs is missing a lot of bits and pieces, but you can much more easily write your own than in XSI (since coding Mental Ray is a nightmare). Perhaps start a new thread for that though? Cheer,s Pete rB Wow that is totally Awesome. I would still like some more nodes for better control over reflections, and specular values in VOPS. I have to admit that I am still not super strong in Mantra Renderman even though I have learned a lot. I am stronger in XSI's Mental Ray Shading Trees. I have a way easier time manipulating the light to surface interaction in their trees because of a few very useful nodes for manipulating Ray Tracing. However, they still don't have the high degree of control over the specular the way they do over the reflections. I could post a more formal layout and definition of nodes if anyone would be interested. There is also a desire for Fuzzy Logic, Bayesian Learning, and Neural Nets VOPS ok now I go run and hide. Would be really useful for POPS, SOPS, and VOPS Shading. lol Cheers, Nate Nesler 23366[/snapback] Quote Link to comment Share on other sites More sharing options...
edward Posted December 21, 2005 Share Posted December 21, 2005 The Polyknit OP still only does Triangles should be an option to do Quads also. The PolyKnit SOP *does* do quads. It just depends on how you do the knitting. Quote Link to comment Share on other sites More sharing options...
Andz Posted December 21, 2005 Share Posted December 21, 2005 Its really detailed, but the modeling Topology is not clean. Quote Link to comment Share on other sites More sharing options...
MatrixNAN Posted December 21, 2005 Share Posted December 21, 2005 The PolyKnit SOP *does* do quads. It just depends on how you do the knitting. 23369[/snapback] Oh wow your right. I just did it. Very cool. I thought there was only one way you could knit. It seems you have to knit right to left when doing the quads or the normals get fliped maybe that was just with that grid and not always. Well I guess I don't have much to ask for in the Quads section except for on the bevel option. Also a Convert Triangle to Quad tool would be good for clean up operations. The L-System Quads would be cool that kind of work like the ZSheres do in ZBrush. Because then you could take your L-System Trees into ZBrush and sculpt them up into high detail models and export the displacement maps and normal maps to be placed on the L-System Tree. Which would allow for more detail and you could create more than one displacement tree per L-system which would allow for more variation with no additional render time. Thats why I ask for it. I guess the only thing else I could think of would be for COPS to get more Shake features in it. Quote Link to comment Share on other sites More sharing options...
MatrixNAN Posted December 21, 2005 Share Posted December 21, 2005 But huh... who models grips on radio buttons?!?!? I bet the screws have threds too... 23370[/snapback] Um well I guess I would be guilty of that. lol XK140 WIP of Windshield Wiper and Mirror Quote Link to comment Share on other sites More sharing options...
sibarrick Posted December 21, 2005 Share Posted December 21, 2005 Hmmm ... triangles are well-defined. Would you prefer quads to triangles even if the quads were non-planar? 23358[/snapback] For organic stuff definately. Subdivided quads even non-planar ones always produce cleaner models. Also if you then have to do further hand modelling afterwards it's much easier if you still have quads. What I'd like to see in the simplest case is that the cookie sop only triangulates the polys it has to modify and it leaves all the rest alone. Ideally it would be able to build quads at the intersections too but I know that is way more difficult. When I was toying around in the hdk doing tri to quad conversions I came across some good algorithms for completely converting tri models into quads but they are pretty heavy coding, so i such with a simpler approach that leaves triangles in tricky cases. Quote Link to comment Share on other sites More sharing options...
Andz Posted December 21, 2005 Share Posted December 21, 2005 Um well I guess I would be guilty of that. lol XK140 WIP of Windshield Wiper and Mirror 23373[/snapback] Oh well... if they are there I guess the guy needed it but. Usually on an assembly file, we only show a simplified part. And apply texture map is the way to go on theses kind of details. Even in CAD. Quote Link to comment Share on other sites More sharing options...
Jason Posted December 21, 2005 Author Share Posted December 21, 2005 Oh well... if they are there I guess the guy needed it but. Usually on an assembly file, we only show a simplified part. And apply texture map is the way to go on theses kind of details. Even in CAD. 23376[/snapback] It's a CAD model from the manufacturer of the car, yup - certainly modeled with patches to engineering detail. This is a polygon conversion and blind export of the car. Quote Link to comment Share on other sites More sharing options...
Mario Marengo Posted December 21, 2005 Share Posted December 21, 2005 Just for fun, check this out. This is a Mantra rendering of ambient occlusion, against ~10million polys. Mantra chewed it up in 9 minutes, doing 32 AO samples per, er, sample. 23360[/snapback] Yup. Mantra got a significant occlusion speedup in H8 -- especially noticeable with polys. Kudos to the Mantra team! Quote Link to comment Share on other sites More sharing options...
MatrixNAN Posted December 21, 2005 Share Posted December 21, 2005 For organic stuff definately. Subdivided quads even non-planar ones always produce cleaner models. Also if you then have to do further hand modelling afterwards it's much easier if you still have quads. What I'd like to see in the simplest case is that the cookie sop only triangulates the polys it has to modify and it leaves all the rest alone. Ideally it would be able to build quads at the intersections too but I know that is way more difficult. When I was toying around in the hdk doing tri to quad conversions I came across some good algorithms for completely converting tri models into quads but they are pretty heavy coding, so i such with a simpler approach that leaves triangles in tricky cases. 23374[/snapback] I agree with this. Totally. Cheers, Nate Nesler Quote Link to comment Share on other sites More sharing options...
MatrixNAN Posted December 21, 2005 Share Posted December 21, 2005 It's a CAD model from the manufacturer of the car, yup - certainly modeled with patches to engineering detail. This is a polygon conversion and blind export of the car. 23379[/snapback] I am surprised its not all parametric Quads if it was converted from NURBS. Cheers, Nate Nesler Quote Link to comment Share on other sites More sharing options...
sibarrick Posted December 21, 2005 Share Posted December 21, 2005 I am surprised its not all parametric Quads if it was converted from NURBS. Cheers, Nate Nesler 23382[/snapback] Just a guess but I bet triangles occlude quicker. This is very interesting stuff actually. I've always wanted to do occlusion renders on our stuff but always thought it would kill the render, maybe it's possible to do now though. Quote Link to comment Share on other sites More sharing options...
MatrixNAN Posted December 21, 2005 Share Posted December 21, 2005 Hey Guys, What do you think about a new section in Houdini Called AIOPS? Broken into 3 menus: Neural Nets, Bayesian Learning, and Fussy Logic with a set of nodes for each menu. That or put them into the VOPS. Are other people interested in this and where do you think they should go? Cheers, Nate Nesler Quote Link to comment Share on other sites More sharing options...
mark Posted December 21, 2005 Share Posted December 21, 2005 Data in Star Trek had a neural net didn't he? Quote Link to comment Share on other sites More sharing options...
edward Posted December 22, 2005 Share Posted December 22, 2005 What I'd like to see in the simplest case is that the cookie sop only triangulates the polys it has to modify and it leaves all the rest alone. Ideally it would be able to build quads at the intersections too but I know that is way more difficult. Hmm ... I just took two torus and ran it through the Cookie SOP. It seems to leave the original polygons to me. BTW, Data had a positronic brain. I'm not sure if they ever mentioned it was a neural network or not. My personal opinion from reading around is that Quantum Computing will at least be needed in order to simulate thinking. Quote Link to comment Share on other sites More sharing options...
thekenny Posted December 22, 2005 Share Posted December 22, 2005 BTW, Data had a positronic brain. I'm not sure if they ever mentioned it was a neural network or not. My personal opinion from reading around is that Quantum Computing will at least be needed in order to simulate thinking. 23388[/snapback] HOLY geeked out madness Batman! All we need now is for Stu to tell us how many frames the wipes are in the Star Wars movies. (again) -k Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.