| Author | Message | ||
     Dave Rating: N/A |
Hey, why not making a Direct3d or OpenGL program that changes all D3D/OGL games into anaglyph stereo image? It's been done, but very poorly. Although, look at Alaglyph quake, it's amazing... I emailed the creater, and he said he might take on such a challange, but he might not have the time, nor skills needed. This sounds like a fun project to do for the Income Impared persons not being able to afford shutter glasses, don't you think? Well anyways, I would realy like some response. Later. | ||
| Michal Husak Rating: N/A |
Hi. I thing that making an anaglyph wraper in the age of shutteglasses is like using stones to fight tanks in 21 century wars. I have studied this probelmatic , the main trouble with anaglyph creation is, that it could not to be as good HW accelerated as interlaced or even HW page flipping surfaces creation. The good converion is very slow. The HW suport with is is optimal for such conversion (as acumulation buffer for OpenGL) is often only software emulated in the graphic card drivers. The drivers you discuse de facto exist, if you use VRcaddy and set anaglyph mode on Winx3d you will get a D3D game in anaglyph. But it will be extremly slow .... You shuld try to ask the people from Brightland, theyr Acumira technology is superior to anaglyph and they have probably found a way how to make it enought fast .... | ||
| Shockster Rating: N/A |
Hey, thanx for the info, but what if you just use 1 colour for each interlaced frame buffer? It sounds easy enough, like modified Revelator or other drivers. The only thing I have to say about the "fighting tanks with rocks" comment, is I can't exacly afford $50 for shutter glasses. And I'm sure there's more than just me that can't, and for us to not have true 3d is somewhat.. cruel.  And as for VR Caddy, it costs money also.. and like you said is very slow. Thanx for everything, cya Dave/Shock | ||
| Brightland Rating: N/A |
Hi guys, Akumira is indeed fast enough for real-time games, and can be 100% hardware accelerated with current graphics hardware. Please take a look at R4D at http://www.brightland.com/sourcePages/r4d_v9.htm . R4D implements real-time Akumira, and is very fast. It could be even faster if incorporated at the driver level. If there is enough interest, we'll approach NVidia, etc., and see if they would be interested in adding support for Akumira. While shutterglasses are the best all around quality for stereo3D, the best you can do on LCD's, flat panels, and notebooks is Akumira, where shutter glasses won't work. If the application does not require 100% color accuracy, then Akumira is a fine choice either way (no flicker, no batteries, many people can watch the same display with cheap glasses, etc.). We're still working on low cost versions of our anaglyph glasses improvements. These new glasses, combined with Akumira, make a viable low-cost, high-quality solution for stereo3D viewing on any display device (including print). More info on Akumira is available at http://www.brightland.com/Akumira/Akumira.htm . Regards, John Schultz Brightland www.brightland.com | ||
| Shockster Rating: N/A |
Hey John, I was just wondering how you are supposed to determine how much interest there is in this technology? I could run a poll on www.stereovision.net just for the hell of it, to see how much interest _I_ could get.. since it's mostly a gaming site, it would be a good site to post it on.. I'm looking forward to see what you can do at the driver level, to imporove the speed of R4D, as if it's not fast enough It's a very nice program indeed, as I've seen it before, but seeing is I'm not a programmer, I don't have what it takes to test, and maybe incorporate that into a game, or direct 3d directly... Thanx again, Dave | ||
| Brightland Rating: N/A |
Hi Dave, Sure, run a poll on www.stereovision.net and see what the interest level is. You might want to ask people to try R4D (or at least look at the Akumira images) so they understand what the technology can do (over standard anaglyph). Please let me know how it goes. Regards, John Schultz Brightland www.brightland.com | ||
| Shockster Rating: N/A |
Hrm, it seem it's split 3 ways. Some people hate the idea, some would useit a but, and some love the idea... there are only 8 votes so far though. I think I'll wait another week or so. Talk to ya later. Dave | ||
| Doon Rating: N/A |
You can't afford $50.00 for glasses? You've gotta be kidding. Just give up the dope, booze, and/or ciggarettes for a week and booya, new glasses. You must be a stoner or somthing if you cant get off your ass and get a job and SAVE $50.00. Your pathetic. | ||
| Alexander Oest Rating: N/A |
I don't think Dave's pathetic. Low end stereoscopy can stimulate the development of more content - movies, games etc. - which in turn will give more high end content and we will all - rich as well as poor - be happy and live together in peace and understanding and the world will be a better place. Alex ...who happens to be filthy rich but has a sick passion for cheap stereoscopic display systems. | ||
| Shockster Rating: N/A |
Excuse me? Pathetic? Dood, I'm 15, and don't have a job, and rather than wasting time at McDonalds making $4 an hour I've decided to develop my skills on a computer. I know you're gonna say 'Then don't play games in 3d' but it's more than games I want this for. I'm a 3d artist, and use 3d Studio max, and have been for a year now, and because I didn't get a job at some deadbeat fast food restaurant, I've become quite the talented person. Http://shockster3d.cjb.net That's just some of my work, I'm getting better, mind you. I'd ask my mom for some money, but she's in a rut right now. And I'd rather give her $50 then spend it on shutter glasses. I'm not a boozer, I'm not a stoner, I'm definitely NOT a crack/coke/speed/'X' head. That does it, I've said too much Hope that clears up the opaque view you had of me Seeyah (Oh and John, seems more than not, ppl want it, what do you have to say about it? Besides, most of the ppl on the forum have shutter glasses anyways) Dave | ||
| Anonymous Rating: N/A |
Hi Dave In this situation you shuld try to do folowing: Download from Microsoft DirectX SDK and study deep. Have a look on Acumira images, you will be able to catch the color transformation they use from the image analysis. Study C,C++ and DCOM technology, you are young you will not have trouble to catch it. If you spend about 3 hours dayly on it, you will be able to write a good working anaglyph game wraper after one year or so and make money on it .... | ||
| Brightland Rating: N/A |
Hi Shockster, Thanks for doing the survey. I can understand those who voted "no" as they have probably never seen Akumira 3D. Adding support for standard color anaglyph would still be cool, but the quality would not be very good. Send me your mailing address and I'll send you some free anaglyph glasses that work well with Akumira. I sent an email to NVidia regarding Akumira support at the driver level... Hi Michal, Akumira is more than a color transformation... Creating Akumira required the development of some fairly sophisticated tools. The Akumira process, and resulting images are unique and patent pending. Thus, copying it would probably not be a very good idea if the goal is to make money selling a clone. Our goal is to provide quality stereo3D at a very low cost to everyone. We do not currently have the resources to convert existing games to Akumira format. Our experience in converting Quake to stereo3D for H3D (which launched the stereo3D market for PC's) was that the right way to solve the problem was at the driver level. There are significant optimizations that can be done at the driver level that cannot be done at the application level. Thus, driver level support will always be faster (much faster than PFD_STEREO as well). The other major benefit is full support for all games and applications, not just a few. Regards, John | ||
| Michal Husak Rating: N/A |
Hi John I am not sure what you mean under support on driver level. If you mean adding stereoscopy automaticaly to non-stereoscopic aplication, this concept is totaly wrong becouse the aplication developer have zero control about scene stereoscopy geometry. Using PFD_STEREO OpenGl is a drivers build-in future as well as the matrix operation used for stereoscopic scene generation and I have no idea why this shuld be slower than other method ... | ||
| Brightland Rating: N/A |
Hi Michal, PFD_STEREO only turns on stereo framebuffer support, providing left and right draw buffers to render into (does not affect anything else). The application must still set up the stereo matrix transformation per eye, and render the entire scene twice (It would be cool, though, if there was an extension that allowed setting up stereo parameters where the driver rendered the scene twice). This means the application must send down all vertex data twice, as well as set all material properties twice (especially slow if textures are being paged). Remember the R4D dino, robot, etc., demos? Using PFD_STEREO in OpenGL on the Verite (way back when), it ran around 15 fps. Using my own custom transform and clipping code and the low level Rendition Redline API, I was able to hit 60+ fps. 1. I was able to optimize clipping and projection as only the x coordinate is different per eye. 2. I only traversed the scene graph once. 3. Materials did not have to change for left/right triangles. Thus, when using Redline, I was writing effectively at a T&L driver level. When Metabyte first showed us Quake running in stereo3D at the driver level, it didn't look nearly as good as our custom port. But, this also had to do with Quake not drawing everything in 3D, which caused some major problems. Later on with tweaking, Metabyte made its driver-level solution look pretty good. It then became possible to play just about any game in stereo3D. It turns out that there is a limited useful stereo3D view volume for pixel coordinates to live in. Depending on where the viewer sits relative to the display and the size of the display, different stereo3D values are optimal. Thus, a developer cannot optimize the stereo3D view volume; it's up to application user to set those parameters. So, given a fixed near and far clipping plane, we can set the stereo3D parameters perfectly for a given viewer's monitor size and viewing distance. How many parameters do we need? Just two! The first parameter sets the near stereo3D depth (out of the screen, typically), the second parameter sets the far stereo3D depth (far into the screen). Once these two parameters are set, you are guaranteed not to tweek the user's eyes, no matter what is drawn. The application programmer cannot know this ahead of time. These two parameters are pixel values. For example, -8 near, +16 far (separation between pixels at the near and far clipping planes). R4D Player uses this technique *dynamically* to minimize ghosting and eyestrain (works well for single objects, but is not recommended for videogames). There is one more parameter that can be used to perfectly optimize the stereo3D viewing volume: a lookup table. This lookup table can be initialized with any linear or non-linear function to completely optimize the stereo3D view volume between the near and far clipping planes. I did this back in 1987 with SpaceSpuds on the Amiga, and it worked great. In summary, driver level support with two parameters and a lookup table will provide perfect stereo3D in any application for any monitor size and viewing distance. The only thing a developer needs to do is make sure they draw the scene using 3D commands. All 2D commands should be placed at the near plane or on the screen surface (or another parameter can be provided as to where to place 2D commands). Regards, John | ||
| Michal Husak Rating: N/A |
Hi John Thank you very much for explanation of your opinion. I realy do know what PFD_STERO does and how the rendering pipeline works + how corect stereoscopic glFrustrum looks like. I am working profesionaly on software for scientific stereoscopic visualisation of molecular data ... I still disagree with your concept. The key for correct stereoscopic visualisation is placing the most importnat objects in the zero paralax plane. Only the software developer could know, what object it is in the given situation. No automatic wrapers could do the job. I agree that optimalization for the screen size and viewer distance from the monitor is useful ... I am just solving this problem becouse different setting is optimal for viewing via LCD glasses and viewing via HMD .... I am not sure how do you want to avoid multiple rendering for correct stereo output creation. Colud you , plese, explain me a bit more precisly this point ? If you use display lists for graphical data storage, you change only the view matrix and all the geometry and texture is in videomemory, but this is an standard common technique. Did you mean this ? | ||
| Brightland Rating: N/A |
Hi Michal, The only way to avoid rendering the scene twice is to let the driver perform the stereoscopic projections, clipping, etc. That's how the Metabyte/Elsa/Asus/Nvidia driver-based stereo3D works. The applications are only sending down the scene data *once*. The driver then takes each triangle, transforms, clips, and projects two triangles (left eye + right eye), which then get rasterized *without* changing the current material state. That is extremely fast, and can only be done at the driver level. Could you do this at the application level? Yes, sort of, but you'd have to change the modelview matrix, projection matrix (you could concatenate them for a slight improvement), and the drawbuffer for every triangle list (however stored) with the same material type. This would save changing material states, but you'd still have to transfer all those verts across the bus twice, which is a *huge* performance loss. Can it be sped up? Yes, you can use Compiled Vertex Arrays (should be called "locked" vertex arrays), or on NVidia-based cards, Vertex Array Range (VAR), which is the absolute fastest method (I use this quite a bit for non-stereo related 3D work). Unfortunately, AGP 4x does not work as designed (nor does AGP 2x), nor do fast writes, so it is extremely important to minimize AGP bus traffic. You still have to send down vertex indices twice (32 bit ints appear to be fastest, even though there is supposed to be a 32767 index limit (16 bit int)). On modern 3D consumer cards, display lists do not end up in graphics card memory (and run much slower than VAR, etc.). The last machines I worked on that had fast display lists (dramatically fast!) was on SGI IRIX-based boxes (Maximum Impact, Reality Engines, etc.). As far as where to place the most import scene elements in the stereo3D view volume: eyestrain is minimized at the zero plane (of course), so I understand your argument to place the most important objects at that location (for small 3D objects, like a cell phone, molecule, etc., R4D places such objects in that exact location, dynamically). Since R4D is a general purpose viewer, I cannot know what the user is going to load into it before hand. So, I dynamically set it up based on the geometry that is present. This is fine for object visualization, but is not optimal for out-the-window games and simulators (which should keep the stereo3D view volume relatively fixed). If a develop wishes to have total control over the stereo3D object placement, they should still allow the user to adjust the stereo3D view volume. Screen size, viewing distance, and perhaps most important: personal preference, will all need to be satisfied. Does driver level stereo3D support limit the developer's ability to place objects in 3D space. Nope. You tune your app for the driver's stereo3D parameters and tell your users what values work best. With a given set of driver-based stereo3D parameters, you can tell *exactly* where objects will be in the stereo3D view volume, giving the developer total control with maximum speed and ease of use. Regards, John | ||
| Michal Husak Rating: N/A |
Hi John Thank you for your explanantion again. The main trouble is in "tune your app for the driver's" . There is no other realy used standard for stereo then the OpenGl one and you do not have acces to the math used in several proprietary wrapers ... I still hope that Nvidia will implement not only the wrapers but a standard OpenGl stereoscopic drivers as well ... Accroding the stereoscopy and speed - I am probably to much focused on profesional 3D Labs graphic card using HW triangle setup engine + display list in videomemory + the SGI HW. | ||
| Brightland Rating: N/A |
Hi Michal, In the consumer market, there will soon be a standard: NVidia's. Once NVidia stereoscopic support is officially released, that will be the standard. For custom visualization apps where you (the developer) have total control over what hardware is used, there is nothing wrong with implementing stereo3D support yourself. If there were a "standard" OpenGL extension for single-pass stereo3D support (set up stereo parameters via extension calls, then the driver does all the work), supporting stereo3D would be much easier, and considerably faster at runtime. I would propose four parameters: near sep (pixel sep at the near clip plane), far sep (pixel sep at the far clip plane), 2D sep (for all 2D commands/primitives) and a lookup table (optional) for interpolated separations from near to far (you can put a curve in the table that maximizes the near stereo3D effect and compresses the far stereo3D effect for optimal volumetric perception, or vice-versa, depending on the application). That's *all* you need for total control. Fast, simple, generalized, understandable, and predictable stereo3D behavior. Regards, John | ||
| Brightland Rating: N/A |
Regarding lookup tables- If lookup tables are used, the driver only needs to lookup using the effective homogenous Z (mapped to 0..1. This is exceptionally easy to use with a lookup table and interpolation between the two nearest table values). Then, use hardware guard-band clipping to clip left/right edges in 2D. Fast and easy to implement at the driver level (I wouldn't be surprised if this is how they are currently implementing it). All developers need is access to near sep, far sep, 2D sep, and the lookup table. In fact, if a lookup table is used, you only need two parameters, the lookup table and 2D sep, as near sep and far sep can be implicit in the lookup table! A two entry lookup table would be the same as near sep and far sep (sort of: this would be straight linear and not "natural" looking). More control (and "natural" looking accuracy) is gained by using more entries, allowing a non-linear effect (for hardware speed, the table may need to be a power of two in size...). Regards, John | ||
| Brightland Rating: N/A |
Lookup tables and user tweaking for stereo: All you need to present to the user is an editable curve/spline and a slider for 2D sep (while viewing a 3D test scene). The driver can compensate for resolution changes to guarantee that the pixel separations will match the user chosen ones at the calibration resolution. Regards, John | ||
| Shockster Rating: N/A |
I just want to ask something... Since the Elsa/H3d drivers render both feilds, and store them in buffer and everything else needed to view stereo, what's keeping someone from taking those drivers, or concept, and rendering the feilds in a red/green fashion? Shockster | ||
| Michal Husak Rating: N/A |
Socker: Making good anaglyph is not so easy. You have to make a bit complicated color converion taking in mid the human eye sensitivity for brightnes of different colors. The anaglyph obtained by the method you mention will be horrible ... | ||
| David C. Qualman Rating: N/A |
As has been mentioned, the WINx3D library does provide anaglyph. And, has been mentioned, it is pretty slow. Mainly because of what Shockster suggested, it runs slow. The wrappers from Elsa/H3D and from VRStandard do render two independent images. But, in the video card's memory. The same would be true if this were done at the driver level. Now, to process these separate images into an anaglyph image, you need to address each pixel from each image inside the video card, compute the anaglyph color, and put the anaglpyh pixel back into the video card. This is a hell of a lot traffic across the PCI bus. Of course there could be optimizations, but it is still a hell of a lot of traffic across the PCI bus. Also, many video cards are optimized for writes into the card, and not reads from the card. The whole process is pretty slow. The best solution is to do the anaglpyh conversion before the color data even gets into the video card (or to have a video card that can do the anaglyph conversion. Ha!). The old game Magic Carpet had an anaglpyh mode. They did the conversion while the image was still in the computer's memory, and not in video card memory. Thus, it could be fast. But not free. | ||
| kgian Rating: N/A |
If you haven't noticed magic carpet has color anaglyph, like Akumira!!! The image isn't monochromatic... Now how they did that back then?? | ||
| Brightland Rating: N/A |
kgian, Simple color anaglyph is easy to do (Magic Carpet, Anaglyph Quake). Color anaglyph that runs fast and looks good (doesn't hurt your eyes) is a different story. Color anaglyph has been around a long, long time (they even tried it on television many years ago). The reason it never became popular was the color quality and viewing comfort were poor. The cool thing about Akumira is that it solves both the quality and comfort problems, and can do so in real-time... John | ||
| Michal Husak Rating: N/A |
There is a relatively easy trick how to create anaglyph fast - change the view and change the color and intensity of all lights lighting on the scene + use normal blending of this two views together ... It colud be done in videomemry, but you need acces to the source code of the game  . According John opinion that "Once NVidia stereoscopic support is officially released, that will be the standard. " it is probably true but I thing that it will be the most horrible thing witch can happen. NVidia position is now equivalent to Microsoft position on the OS market ... There is a risk that they will dictate the standards to the others .... And theyr graphic card are not good, they are only cheap ... Some examples: 1) No HW stencil buffer for 16 bit modes 2) Low quality RAMDAC producing ghosting on hi-quality monitors 3) No support for standard OpenGL stereo even on "profesional models" I can continue a bit longer if anybody interested . | ||
| Brightland Rating: N/A |
Hi Michal, If you have issues with NVidia, why not write to them and ask for features you want? As far as RAMDAC choices, that is up to the board vendor to choose. I've got NVidia cards in almost every workstation (except the servers) in my office, ranging from a GF1 DDR, GF2 MX, GF2, GF2Ultra, (even a TNT2), etc. They have different RAMDACs, and different monitor types (CRT, Aperture Grill, Flat Panel). They all have excellent video output, with no ghosting. In fact, my main workstation uses a Creative GF2Ultra on an 18" NEC LCD1810 flat panel, with an *analog* interface. It is razor sharp, and as sharp as digital interface LCD flat panels. The NEC LCD1810 is about as high of quality as you can get... My home machine runs a GF2 on a ViewSonic Flat Panel (analog), and it too is absolutely razor sharp. The trick to getting a sharp image on an analog flat panel is to go in and fine-tune the settings manually. If you want to spend the bucks, get a digital-out card, and a 1600x1200 flat panel. There are also CRT's with digital in now; you can probably go up to 2048x1536 with zero ghosting... The only reason I see that there is no standard OpenGL stereo (doesn't Elsa support it?), is lack of demand. So demand it! It would be great if 3DLabs and ATI can come up with competing products with NVidia... I kept a 3DLabs GMX2000 (Dual, w/96MB RAM), in my main workstation for a very long time, as it was very stable and fast. It didn't get pulled until the GF2 came out and NVidia's drivers were very stable. Today's video cards are reasonably fast; they now need to work on a better bus interface between the CPU and GPU (put them both on the SAME bus, like "ancient" Amiga, or uhm, what's that new machine, XBox). AGP is very slow... This would require a fundamental change to the PC architecture. Perhaps NVidia will work with AMD and Intel to develop a new standard... This could happen; take the XBox chipset design and make a PC version. Regards, John | ||
| Shockster Rating: N/A |
Wow, thanx guys, I didn't even expect to get 1 response Still though, keep in touch, I want to see if there is any news in the anaglyph scene. Say, is there a dedicated webpage for anaglyph games?? Cheers Shockster | ||
| Michal Husak Rating: N/A |
Hi John We have got the ghosting problem with almost all Nvidia graphic card on market with 22' flat analog monitors (IYAMIA , Mitsubischi ... ) this monitors use to high internal frequence on witch the Nvidia chipset produce ghousting-frequence signal ... It is not only a simple RAMDAC problem ... I am sure that both monitor manufactures and Nvida well know about this problem ... According the existing graphic card design - the AGP solutin is evcose something subobtimal. The solution is unified memory architecture as used with the SGI PC with Cobalt graphic card ... Unfortunatly this solution did not make it ... |
