vvuk / gecko-dev Goto Github PK

Visible in this example:
http://mozvr.github.io/cursor-target-test

Hi,
With the 0.4.3 Oculus SDK, I attempted to link up the Linux build with the oculus branch on this repo. It seems like there were some API changes on the Oculus side, which I think I managed to fix. Right now I'm segfaulting after clicking something like the VR button on the quake test page, or just loading up a webvr enabled page. It looks like currently it segfaults at /gecko-dev/dom/vr/VRDevice.cpp:286 due to mParent being unallocated I think. Here is the traceback: http://pastebin.com/z5pJ2Mgt

Anyways, I've got about a 10 line diff to get it this far which I'd be happy to PR if it would help.

Otherwise, is there anyone working on adding Linux support? I'd be happy to test things out.

Thanks!

hi,

I'd love to check out how this project, but I can't figure out where to download a compiled binary. I'm an OSX user, so preferably a .app. I could only find an old build that was pre DK2 SDK.

Any pointers would be very appreciated!
Francois

Hi Vlad

When I clicked on the link to CSS samples from your latest blog post, I got a 404 on the following URL

/~vladimir/vr/simple-css-2-.html

The DK2 has a built in latency tester. The SDK has an API to access it: ovrHmd_ProcessLatencyTest

https://developer.oculusvr.com/forums/viewtopic.php?f=55&t=10981

Can we expose it in the browser?

In windows, we had fullscreen automatically jump to the correct desktop. This meant that it was impossible to accept the fullscreen without unplugging the oculus display.

Will be awesome if D3D11.1 stereoscopic rendering mode can be added as a VR backend.. as similar to Oculus and other HMD requires two views all work done for VR can be also useful for stereo rendering..
also as Firefox for Windows uses ANGLE which uses D3D11 backend already wouldn't require much work..
D3D stereo rendering requires Windows 8 and up and all current GPUs and "consumer" stereo displays such as HDMI 1.4 3D TVs for example support it.. as said AMD,NV and Intel GPus support it.. and via this API also Nvidia 3D Vision monitors are supported and AMD HD3D monitors..
For a simple example using Stereo API see:
http://code.msdn.microsoft.com/windowsapps/Direct3D-111-Simple-Stereo-9b2b61aa

Implement CSS property to specify that a element is to be interpreted as a stereo left/right pair.

This will be used in cases where we want to create stereo rendered output outside of using requestFullscreen.

On OS X (10.9.2) and DK2 it seems to detect the HMD + orientation device correctly:

2 " VR Devices"
"Found a HMD" PositionSensorVRDevice { hardwareUnitId: "uknownHMDInfo-0x13007a000", deviceId: "somedevid", deviceName: "HMD Position Device" }

However orientation data does not seem to get initialized/updated:

var state = vrPosDev.getState();
console.log(state.orientation.x,state.orientation.y,state.orientation.z);
// => 0,0,0,0

I don't know if this is related, but https://github.com/benvanik/vr.js/tree/master/experimental/usb-driver talks about how

a codeless kernel extension is required to disable the system HID drivers from taking control of the Rift and preventing its use [in Chrome*].

*) Not sure if this only applies to Chrome or if this would affect Firefox as well.

I created this planetarium demo to test performance:

http://swimminglessonsformodernlife.com/cssVR/perfTest/

1. You should see a star counter in the middle of your field of view (press ‘z’ if you don’t to zero head orientation)
2. The number of stars starts at 100. Head tracking works well.
3. Press ’s’ to add 100 stars. Observe how performance degrades. On my machine, the demo is already unusable with 300 stars. Passed 1000 the browser hangs and I have to kill it because I cannot even close the tab.

I’m on OSX 10.9.4 - Macbook PRO Retina 13” - 8GB RAM - 3GHz Intel Core i7 - Intel HD Graphics 4000

Not sure if this is still the place to log issues with Firefox WebVR but I seem to having a bit of trouble with the latest Nightly (43.0a1). I get this on the Oculus DK2 in the sechelt demo on http://mozvr.com/ when I go into VR mode:

Running on Windows 8.1. This was checked with Oculus Runtime 5.0.1 and 6.0.1 and had the same results. I wasn't experiencing this prior to v43. I get a similar thing with my project.

I'm probably missing something obvious but I'm having trouble getting simple WebGL content to display in this FF build. On the get.webgl test page, the cube doesn't display. Running Windows 8.1 with up-to-date graphics drivers; no WebGL issues in FF 33, aurora & nightly.

Currently, all CSS VR content share the same 3D context. This can be a problem in cases where we want elements to always render on-top regardless of their spatial positioning (ie. for HUD elements).

One solution could be to use the css z-index property.

As demoed here: http://leapmotion.github.io/VRCollage/

White background when not fullscreen, purple background on fullscreen. mozRequestFullScreen( { vrDisplay: vrHMD } ) is called on document.body.

Hi, recently I've been adding support for WebVR in OSGJS but it seems like WebVR isn't working on Windows 8.1 with your 33.0a1 build.

(Here is the hmd initialization in OSGJS)

Windows 7 64bits and Mac 10.9.4 have been tested and works fine with both yours and toji's chromium build.
On the windows 8.1 laptop, WebVR is working in the Chromium build but on the FF build, mozGetVRDevices passes an empty array to the callback, not even a fake device.

navigator.mozGetVRDevices( function(){console.log(arguments) } )

window.addEventListener("hmdconnected", function(e) {});

Similar to the gamepad API.

window.addEventListener("gamepadconnected", function(e) {});

We will support HMDs other than Oculus Rift DK1. We need a way to query for the display resolution

I believe this is to spec, from this example: http://www.w3schools.com/cssref/trycss3_transform-style_inuse.htm

However, it does cause an issue. Now that preserve-3d is relevant to both 3d transformations and the VR effect, it means that any WebGL content cannot be rendered in front of a transformed element.

With a wholly WebGL VR effect, an alpha 3d context could be given a unit preserve-3d as well, allowing it to appear in front of transformed dom elements. (Although granted, the manipulations for two matching VR Effect divs would be pretty difficult).

Hi, just tried out the cubes example with DK2 and http://people.mozilla.com/~vladimir/vr/3/firefox-34.0a1.en-US.win64-x86_64.zip
This worked fine, however, all X3DOM content (e.g., http://examples.x3dom.org/simpleExamples.html) only renders once or directly crashes the Browser. In the last version (33), x3dom worked, but of course there was no DK2 support. Tried some debugging but ran out of ideas, esp. since this example (http://get.webgl.org/) also doesn't work... :/

Right now is returning [0,0,0,0] for the orientation.
It would be better if it returns the identity [0,0,0,1]
It would not interfere in the composition of rotations in the application code.

The following doesn't seem to be possible:

<div class="fullscreen">
  <div class="threed content">
    html/css 3d content
  </div>
  <div class="hud">
    non-3d debuging/HUD info or WebGL content
  </div>
</div>

My assumption is that content that hasn't been given preserve-3d wouldn't be passed through the FOV projection, but it appears that it is:

The debug text in the above image does not have preserve-3d set- it is a direct descendant of the top-level container that was taken fullscreen.

getVRDevices returns device information that is not very useful:

HMDVRDevice { hardwareUnitId: "uknownHMDInfo-0x18545800", deviceId: "somedevid", deviceName: "HMD Device" } 
PositionSensorVRDevice { hardwareUnitId: "uknownHMDInfo-0x18545800", deviceId: "somedevid", deviceName: "HMD Position Device" }

The Chromium implementation actually returns "Oculus Rift DK2" as the deviceName.

This information (e.g. DK1 vs DK2) is useful when we need to account for different resolutions and optical warp factors.

Although, ideally I suppose we'd be sniffing for capabilities instead of particular devices in accordance to the web's best practices. Right now my application is guessing whether the user is in a DK1 based solely on window resolution when in fullscreen.

Virtual reality in the web is part of the concept. So thought would share. Being a web developer, a major part of my day goes into inspecting and tweaking elements in the browser. Wondering if the same kind of inspecting and tweaking can be applied to 3D worlds or maybe even the real world. I've made a video to explain the concept. Please find it here: https://www.youtube.com/watch?v=ITvAYepvhN0

Hey, I was testing your new build and it seems like there's a problem with firefox vr distortion.
These demos were working with your precedent build but are now broken.
http://tyrovr.com/lab/webvr/renderer/
http://media.tojicode.com/webvr-test/
http://osgjs.org/examples/dualcam

See for example, tyrovr demo:
Old build:

New build:

Your demo seems fine though

Hi,

when using the getEyeParameters().renderRect size to set the size of the WebGL canvas, then Firefox renders the WebVR output wrong (Firefox Nightly 47.0a1, Windows 7, Oculus Rift DK2, SDK 0.8.0).

The VR rendering is currently only correct when using the screen/window size for the canvas size, but this is WRONG - because this way the VR rendering resolution depends on the resolution of the normal desktop screen - and this one can be lower or higher than the one from the VR device (which could cause too low rendering res for VR or waste of rendering power when higher).

You can see the problem here in the Quake 3 WebGL Demo from Brandon Jones:
http://media.tojicode.com/q3bsp/

or here in the krpano WebVR example:
http://krpano.com/krpanocloud/webvr/index.html?v=119pr3

See also the WebVR spec and example there:
http://mozvr.github.io/webvr-spec/webvr.html#dom-vreyeparameters-renderrect
That means when implementing WebVR by following the specs, Firefox currently behaves wrong.

For the moment I will add a special workaround for Firefox in krpano, but it would be nice when this would get fixed soon.

Best regards,
Klaus

See this video (sorry, it's not the greatest):
https://drive.google.com/file/d/0B7cqxyA6LUpUR25mWjJUc2stclU/view?usp=sharing

I'm applying a css 3d transform. At certain values (which are still valid css), the image disappears altogether.

These disappearances roughly correlate with where the element would have gone entirely outside the window bounds when non-fullscreen.

Near the end of the video, you can see firefox hang & jump for a moment as well. This happens regularly.

Edit: live demo: http://leapmotion.github.io/VRCollage/

Right now all VRHMDDevices share the same HMDInfo, which means that if one changes the FOV (or other settings), they all get the new settings. This is not good.

One way to go around this is to give each VRHMDDevice a callback before it goes fullscreen, so that it can configure the HMD appropriately.

...while the rest remain in the current state.

Something on the lines of the presentation API:

http://webscreens.github.io/presentation-api/