Posts by tksharpless

    Having followed the barebones splat tutorial, using a splat centered in a room, I now have two problems.

    1] There is no mouse or keyboard control of viewpoint. The view simply rotates around the splat center and zooms, just like a spherical panorama. Setting the ox, oy, oz variables in the xml does produce the expected viewpoint shifts, but the single point perspective remains.

    2] several incorrect dark patches on the walls, which are only weakly visible in the Supersplat player, are much darker in krpano. I would guess the difference might have to to with spherical harmonics. Of course this is really a defect in the model, which must ultimately be corrected in the training pipeline, but I wonder if anything can be done in the xml to minimize it.

    Very impressive, model and package. I'm amazed that you can provide so much control of the viewing experience. What are your thoughts about the big problem of keeping the viewpoint from going out of the 'sweet volume' of the model?

    I hope krpano will soon support spherical harmonics for spat rendering. This is a very important element of the superior image quality of splats vs 3d models, because it enables lifelike motion of highlights. Please compare the highlights on this gold-covered statue as rendered by the SuperSplat player, with krpano's current rather dull rendering. SuperSplat version: https://3d3sixty.com/tksharpless.ne…tml?noanim=true . Splat data to drop into the krpano demo: https://drive.google.com/file/d/1VeeEky…?usp=drive_link

    Thanks, Tuur

    I'm having trouble understanding three things: how to adjust the internal coordinates of a splat (which as far as I know are established arbitrarily by whatever tool created the source camera alignments); how the scene coordinates of the splat are generated from its internal coordinates and the krpano parameters; and how those scene coordinates affect display perspective and navigation.

    What I really want is 'object movie' style controls, that let you orbit around the model without experiencing extreme perspective distortion. I don't think there is anything like that in the 1.23 demo; or if there is I have not been able to find it. But I have several times encountered extreme perspective distortion.

    I've been able to build with PostShot a Gaussian splat model of a sculpture, as an isolated object without background. When dropped into the krpano test page, it looks very good, but also very remote and initially hard to find. And once found, the viewing controls rotate it around some far distant point, so it is not easy to see all sides. I tried another splat file exported after positioning the object so that the PostShot coordinate origin was inside it, but the result was some kind of exploded view looking nothing like the object.

    So I need advice on how to make a krpano page with such a splat, that will behave like an object movie: the object initially centered, and easy to fly around. I have not made any splat pages yet, so please assume perfect ignorance.

    I'm looking for information on splat file formats, especially for examples of real, working files, in the formats that krpano can load, but interested also in others.

    My aim is to create scripts for exporting models from the triangle splatting engines, that will be close to what krpano is already comfortable with.

    In fact I have a notion that, pending Klaus implementing native triangle splats, it may be possible to convert triangle splats into bogus Gaussian splats that will play on krpano 1.23.

    So, please, if you have a .splat or .ply file that works with krpano, please drop me a link.

    I've been quite impressed by the triangle splatting project (https://github.com/trianglesplatting/triangle-splatting) and I'm wondering if you will consider also supporting that way of displaying radiance fields?

    This method creates a 'soup' of 3D triangles, each with a spherical harmonic radiance field, that can supposedly be rendered more efficiently than a Gaussian 'soup', and with better visual fidelity. The training process definitely runs faster than for a Gaussian splat, I can do it in 30 minutes on my little gaming laptop.

    Despite the title of their paper, that team has not produced a usable viewer, and their native renderer runs much slower than real time. It is possible to build one-off viewers by exporting their triangles to a game engine, so far only via the primitive .off file format, with fully opaque triangles. However if you would like to experiment with rendering triangle splats, I would be glad to help by implementing exports in other formats.

    Thanks,

    I'm developing a vtour skin with better support for stereoscopic tours, based on krpano 1.22.4. One important feature is to enable side-by-side 'fake VR' display on all flat screen devices, so that the user has the option of using a stereoscope.

    It is reported that on 18.3.2 iPadOS my skin does show the VR button, but that tapping it does nothing. Unfortunately I do not have an iPad for testing. If you have one, please visit https://tks-vr-tours.s3.us-east-1.amazonaws.com/BNJ-OnDeck/tour.html and try for yourself.

    I would not expect any problem with side-by-side display on an iPad, so would assume that my code has somehow disabled that. However I am puzzled. All references to this button in my skin xml have to do with its visibility; the webVR plugin, which I have not modified, is responsible for reacting to it. My xml has   mobilevr_fake_support="true", I can't see anything in webvr.xml that would explain the problem.

    Is it possible that the person who reported this has somehow configured their iPad so that fake VR fails?

    The so-called 'fake VR' display mode is very useful for stereo enthusiasts as it allows viewing panoramas with a stereoscope -- not only 'cardboard' style phone accessories but also large format professional units. For the latter, an iPad is an ideal display device, but unfortunately by default krpano does not enable 'fake VR' on iPad, even when it is enabled for phones and desktop. Can anyone show me how to correct this?

    SOLVED. The correct solution, as suggested by indexofrefraction, is to keep the renderloop running as long as webvr is enabled. It posts its findings to global variables, and the consuming code is triggered by onchanged(var, action). The loop posts a variable only when it is zero (and other conditions are right), and the consuming code ignores the value zero.

    This way gives the effect of cooperating threads, even though krpano runs on a single thread.

    This has enabled me to create a new skin for stereoscopic tours, that works to my liking on all devices including VR headsets. Example here.