Forum Replies Created

Viewing 15 posts - 316 through 330 (of 480 total)
  • Author
    Posts
  • in reply to: Complete Installation Instructions #101604

    Oliver Kreylos
    Keymaster

    Yes. The file name needs to be BoxLayout.txt , or you can name the file arbitrarily and place it into an arbitrary location, and pass its full path and name to SARndbox via the -slf <sandbox layout file name> command line parameter.

    in reply to: Complete Installation Instructions #101603

    Oliver Kreylos
    Keymaster

    Did you add -fpv to the command line?

    in reply to: Some errors at startup #101582

    Oliver Kreylos
    Keymaster

    SWR is neat, but the 29x – 52x speed-up can not be expected for this application. The primary difference between SWR and vanilla Mesa is multi-threading in the vertex processing stage, meaning that SWR can use multiple CPU cores for geometry-heavy workloads like the isosurface/slice rendering for large 3D data sets they are targeting, i.e., tens of millions of triangles per frame. The 29x speed-up for 3M triangle geometries is achieved on a machine with 36 CPU cores.

    Furthermore, the AR Sandbox is fill-limited, not geometry-limited. The rendered surface consists of exactly 612,162 triangles ( (640-1)*(480-1)*2 ), and the majority of work is per-pixel work in the fragment shader. The water simulation is even more extreme; it only renders 2 (two) triangles per computation step, and everything is done in fragment shaders.

    The fragment processing stage of SWR is not architecturally different from vanilla Mesa, which is already multi-threaded.

    SWR is meant for one specific application: visualization of tera-scale 3D data on massively parallel supercomputers, using fixed-function OpenGL as employed by Vtk/ParaView/VisIt. It probably won’t help much on a typical gaming PC or laptop with 2-4 CPU cores, and with fill-limited software using complex shaders.

    in reply to: Some errors at startup #101581

    Oliver Kreylos
    Keymaster

    Intel Corporation Mobile GM965/GL960 Integrated Graphics Controller

    That’s not going to work. Intel’s GMA X3100 graphics adapter, which is based on the GM965/GL960 chipset, is from 2007, before several graphics features required by the AR Sandbox even existed. When SARndbox requests those features, the Mesa OpenGL library falls back to software emulation mode, which is not only orders of magnitude slower than hardware rendering, but also quite buggy, especially in the modern features that don’t make sense in software emulation, and are therefore hardly used or tested. My guess is the crashes you are getting are happening inside the Mesa library itself, not in SARndbox.

    Regarding speed: I just tested the AR Sandbox software on my new laptop, which has an Intel Core i5-5200U CPU with integrated HD 5500 graphics adapter. This is a 2015 CPU, and still doesn’t support the framebuffer / shader features required to run the water simulation. Without water simulation, the sandbox ran at a rate of around 25 frames per second, with visible glitches in contour line rendering (which hint at graphics driver bugs).

    According to the wikipedia article I linked above, this GPU is on the order of 50 times faster than the GMA X3100 (based on comparing core clock rate times number of computing units).

    in reply to: Some errors at startup #101556

    Oliver Kreylos
    Keymaster

    The (lack of) output is a good data point.

    What GPU (and driver) do you have? Please run

    $ lspci | grep VGA

    and

    $ glxinfo

    in a terminal and post the results here.

    in reply to: Minor software update #101540

    Oliver Kreylos
    Keymaster

    You won’t have much success with that machine. Your CPU (Athlon 64 X2) is about a factor of 7 slower than our minimal recommended CPU (Intel Core i5), and while I didn’t find a Radeon card with a 1540 or 1520 model number, I’m assuming it’s the same vintage as the CPU (2009), and probably won’t be powerful enough.

    You might be able to run the AR Sandbox without water simulation by appending -ws 0.0 0 to the command line, but even then it might be slow.

    To avoid the error message during calibration, ensure that the Kinect tracks your calibration target properly before confirming a tie point, and that you don’t move the target for 2 seconds after confirming. The target disk is supposed to show up green in the real-time calibration display.

    in reply to: Some errors at startup #101539

    Oliver Kreylos
    Keymaster

    The required command line parameter is -fpv (“fix projector view”) not -ftp. This shouldn’t cause an issue, but could you please try again?

    Also, what is the full output from SARndbox when it crashes?

    in reply to: Common Issues (READ THIS FIRST!) #101533

    Oliver Kreylos
    Keymaster

    Raspberry Pi natively only supports OpenGL ES (OpenGL for Embedded Systems). The AR Sandbox software is not compatible with OpenGL ES. You can in principle run the AR Sandbox on a Raspberry Pi by installing the Mesa OpenGL library, but performance will be very poor as Mesa emulates OpenGL in software running on the Pi’s main CPU, which is very weak.

    in reply to: Common Issues (READ THIS FIRST!) #101532

    Oliver Kreylos
    Keymaster

    The ProjectorMatrix.dat file is created by the projector calibration step (step 10 in the complete installation instructions).

    You can run the AR Sandbox without projector calibration by leaving off the -fpv command line parameter, but then you will get a regular 3D rendering of the scanned sand surface which will not match the physical sand surface (but which you can pan/rotate/zoom using mouse and keyboard).

    in reply to: Complete Installation Instructions #101517

    Oliver Kreylos
    Keymaster

    I see. The AR Sandbox as it is might not be the ideal tool for that, but the next major version will have a mode where you can pre-load an existing 3D topography model (such as your school grounds) and an aerial or satellite photo, and the software will guide you towards re-creating that model in the sand, and then project the image onto it once the real sand surface matches the model.

    in reply to: Can't edit old posts? #101511

    Oliver Kreylos
    Keymaster

    I fixed one link. Is there another?

    in reply to: benq keystone #101509

    Oliver Kreylos
    Keymaster

    Please see my reply above.

    in reply to: benq keystone #101508

    Oliver Kreylos
    Keymaster

    Yes, I did indeed misunderstand. Just to be clear: When you turn on the projector, it will apply a (bogus) keystone correction setting, which will result in a squished trapezoidal image, and you have to manually undo that every time by resetting keystone correction to zero? Furthermore, once you do reset keystone correction, if you bring up a test image (run XBackground from the Vrui package in full-screen mode), the horizontal and vertical lines have no jaggies, the circles are round, and the horizontal bar in the center has clean alternating black/white vertical lines?

    In that case, you’re probably running into an automatic keystone correction “feature” that’s new to this projector line. None of ours are doing this. Turn the projector on and go to the “DISPLAY” menu. See if there’s an entry for “Auto Keystone.” If so, make sure to turn it off (see page 23 in the projector manual). That ought to fix it.

    in reply to: Lava #101491

    Oliver Kreylos
    Keymaster

    Please see Minor Software Update. I added a graphical user interface to control fluid behavior, and the water attenuation slider makes for maybe not physically accurate, but visually compelling lava at higher settings (around 0.95-0.99).

    Getting the exact behavior you want would require relatively big changes to expose bedding friction in the simulation. I can’t do that right now.

    in reply to: benq keystone #101490

    Oliver Kreylos
    Keymaster

    Have you tried a projector calibration without keystone correction? If you haven’t, would you mind doing it, to see if it works?

    You will have full overprojection during setup and calibration, but once you run the SARndbox itself, the image should be black outside the sandbox’s extents, for the same effect as using the projector’s keystone controls (which don’t do physical blanking either, simply set outside pixels to black).

    You should only get overprojection if your Kinect captures surfaces outside the sandbox, such as the floor.

Viewing 15 posts - 316 through 330 (of 480 total)