Fool me, this is real good shit ! [the VR thread]

[gravy]

ah, wasn't aware of that. I thought time of flight were generally too pricey.

It kind of sucks for the way Kinect was designed, you get a much higher rez zbuffer image but it's nosier than a structured light one. So you need to smooth out that image over time which adds noticeable latency to stuff like skeletal tracking. For this kind of appplication you just want the zbuffer and I reckon that could be done in a low latency becuase the you can't move a massive distance the room you're in would mostly be static.

My immediate take away from the announcement is they have a lot of what they need to make something completely aweseme. The technology they're saying they want to use isn't science fiction.

So it's all down to execution and packaging it into a device that's user friendly and doesn't cost the earth.

[Dio]

The Kinect technology experience is massive for this sort of stuff.

[gravy]

Pinning looks good (I can see a little lag) but it's all overlay, no masking as far as I can see

[F-ZeroOne]

So it's all down to execution and packaging it into a device that's user friendly and doesn't cost the earth.

Pick two.

[pogo]

That’s very interesting, though it’s a bit of a con. You see a short section of what she sees which looks very rough; a little lag, colour is completely off, and some artefacts in rendering. Looks e.g. like the edges of his trousers are imperfectly marked, due to blue screening perhaps. Would be interesting to see how that was captured, to get the views from all sides (or at least two sides). After that though it’s not what she’s seeing but being captured by another camera. That’s not a norma camera at all. Possibly it’s the current prototype or a variant of it, which Verge described:

It's a contraption, to be sure. There's a small, heavy block you hang around your neck which contains all the computing power. It's comprised of lenses and tiny projectors and motion sensors and speakers (or something that makes sound, anyway), and god knows what else.

That’s being held very steady to minimise (though not eliminate) lag. Also being largely stationary minimises the complexity of what it’s doing.

It's also fake another way: how are they capturing video of what she’s seeing? They’d have to replace one of her eyes with a camera ! More plausibly there’s a camera recording what the device sees which is being composited with the “hologram” on another device not far away. But given the lack of a ‘contraption’ hanging around her neck or even wires all of what she’s seeing must be being rendered elsewhere. So even the first bit which supposedly shows what she sees must be faked.

And it shows the other problem with it; how many of us have furniture that looks like that, highly geometrical, heavily textured, otherwise empty? It will be much harder to do such pinning in a real world environment with more complexity and clutter.

[Darkflame]

On the contrary, more clutter = easier to track points.
It makes occlusion harder, but a clean white room is far harder to track then a heavy varying environment.

Did you see the tablet video I posted;
https://www.youtube.com/watch?v=H2VFSo6 ... 11374#t=48
You can download that app right now and use it.

While of course they are going to show it off the best they can - theres nothing impossible or needs to be faked here.

Also, the CNET video clearly shots the camera the audience sees though, it is big and bulky but they arnt hiding it or "faking". You cant film though HMDs lens, so you need a alternative set up to give a audience video.

They’d have to replace one of her eyes with a camera

Or its just a cm or so above or to the side of the eye like existing HMDs have. You cant really tell the difference.

---
Anyway;

Yeah, this is looking fantastic.
Even with expensive cost, lag, and some other issues. This is Dennou Coil level AR a decade early.

Doesn't matter now really. This is good enough to prove its awesomeness to sceptics - investment will flood in even more then already. (Vuzix got a massive investment just a few weeks back).

So, yeah, I'm hyped. This is "me" winning. This is the child me a long time ago finding out LCD displays were transparent and then drawing a picture of two of them stuck on a pair of glasses (seriously, I did that, even called it "skywritter" was basicly a "gameboy in the air". I didnt understand anything of focusing problems though, all youd see is black smudges....still...point is I have dreamed of this for awhile).

I could even hyperbole here about how long term this can help humanity. Using more virtual products rather then physical displacing a fair bit of our ecological footprint. "ideal" AR specs can replace a lot, and I now and very sure we will (eventually) get them.

[/hypeing fanboy mode off]

Open questions (I think):

1. What does this "bounces light around millions of times" mean? why? Whats it doing? Is it to get focus right?
2. How does it work as a platform?
2b. Can multiple separate apps run at once?

[pogo]

Ben Kuchera clearly didn’t get an invite for a private HoloLens viewing, and probably won’t get one after this:
Skeptical of HoloLens? It's time to rewatch how Microsoft sold us on the Kinect

That hardware, as shown, never really existed. In practice it had trouble tracking multiple people. It often lost track of your body, and voice commands were often hit and miss. The Kinect's long journey, and Microsoft's attempts to turn it into a mainstream product, came to an end when it was removed as a mandatory pack-in with the Xbox One hardware.

The Kinect technology is now an interesting memory and, while some people still use in their home, there are precious few games for the device and performance remains hit or miss. Heck, according to the in-depth Wired piece this is a branch off the Kinect tech-tree anyway.

We're not saying the HoloLens is just smoke and mirrors, it's just worth taking a step back and realizing that what they're showing right now is a huge leap from any technology that has existed before. How the hardware will eventually work in our homes, and at what price, are still open questions. Also, heck, it could be smoke and mirrors.

[Darkflame]

I'm reading hands on from everyone else.

The developer system - which real people outside microsoft have used and played about with - has its computing power around the neck.
But it does work;

HoloLens systems will be self-contained and cord-free, the developer units we used had a large compute unit worn on a neck strap and an umbilical cord for power. Production hardware will automatically measure the interpupillary distance and calibrate itself accordingly; the dev kits need this to be measured manually and punched in. The dev kits were also heavy, unwieldy, fragile, and didn't really fit on or around my glasses, making them uncomfortable to boot.
But even with this clumsy hardware, the experience was nothing short of magical.

Through it all, the 3D effect was thoroughly convincing. The system felt very low latency; as I moved my head and walked around, the objects retained their positioning in the real world, with the castle, for example, never becoming detached from or wobbling around on the table. While Minecraft of course falls some way short of having photorealistic graphics, the melding of real and physical nonetheless felt convincing.

http://arstechnica.com/gadgets/2015/01/ ... tual-real/

This ones good because it takes a better stab at technical details.

right now is a huge leap from any technology that has existed before.

Not really, no.
Its combinations and refinements on whats seen before. If they can sale it for a not crazy price in a year timeframe - yes - that's amazing.

But whats shown isn't a huge leap from things available right now, and stuff vastly smaller groups like SpaceGlass have made. Big wallets and lots of experience of microsoft engineers should produce better results then what "a dozen people and a kickstarter" type things can pull off.

Also, Markerless tracking from 2007;



With marker, but a rather cool demo from 2009;

(Non-rigid surfaces is seldom attempted even today)

There simply hasnt been great investment in this stuff till recently, but "amateurs" and researchers have been doing the necessary components for a long time.

[benedictw]

Took a closer look at that video from the previous page :



On this screenshot (click to enlarge) some of the floating elements (RHS) are darker than the background. If thats the case (and not some clever hack) then they've really got something here. LCD + Holographic Lens, perhaps?

Looks like interaction is the 100lb bugbear though, some sort of 3d wand with a button would be better than the fumbling we see.

[gumboot]

On the contrary, more clutter = easier to track points.

Quantity is not quality.


But I'm going to keep on saying it until somebody listens.

Humans implement their image stabilisation mechanically, and they publish the data real-time right out the front of their faces on orbs marked with high-contrast fiducial patterns. Whenever it does glitch their brains blank out the effects. It may well be the most reliable, low-latency contlibutor to the filter network.


If I understand correctly, artificial-eye-camera operators don't have that muscular control, so even they can't offer the necessary stabilisation that could give a perspective that would make sense. Even then viewers wouldn't have the subjective blanks around saccades, etc., so really any recorded demo is doomed to being completely artificial and unrepresentative.



I don't know how wide a structured-light Kinect is, but it seems reasonable to imagine it's not much wider than a human head, so mounting the projector on one side and camera on the other of a HMD seems like it should work if you needed it to.

I think TOF might suffer from needing to throw out a lot more power. Rather than concentrating energy on thousands of points it has to get some measurable response from continuous surfaces (more or less) -- some of which might be quite dark or pre-saturated in IR -- and to be able to make an analogue measurement of that response rather than just a binary "is there a point in this pixel?". Oh, and eye-safety puts a cap on that, too.

I keep thinking of the sights in the Predator movies... do you suppose they considered eye saftey?

[gravy]

This is the XBOX ONE TOF camera they're already using

You can see the noise increasing at the edge of frame though I'm sure I read the Holothings camera has a wider FOV

[gravy]

Gizmodo hands on report, a few more details in here about the current state of the kit and the demo environment

http://gizmodo.com/project-hololens-han ... 1680934585

[pogo]

On the contrary, more clutter = easier to track points.
It makes occlusion harder, but a clean white room is far harder to track then a heavy varying environment.

They’re hardly using a clean white room; a textured, shaded cuboid for the one supposed demo of pinning (though I noted it was clearly not done by the thing she was wearing but on a nearby PC).

My point about clutter though was that it will be much harder to pin objects which are already covered with objects. As e.g. every flat surface I can think of in this house is. Even floors are covered in furniture and depending on the room other clutter. It will look badly broken if your Minecraft for HoloLens sits on your coffee table but with the vase, fruit bowl sticking through. In fact and even worse it's very unlikely to do the calculations to intersect them properly so more likely they will be partly obscured but appear at the edge but looking very wrong as the distance they’re at coincides with what the glasses are drawing.

[gravy]

Why's anyone bothering to debate this with someone who doesn't even understand the difference between motion capture and 3d scanning?

[pogo]

http://gizmodo.com/project-hololens-hands-on-incredible-amazing-prototy-1680934585

Even more detail than the Verge report:

As real as Microsoft's hefty prototype would allow for, anyhow. Saying it's a work in progress would be an understatement. The final version is supposed to be a small, light, tether-free set of glasses you wear on your head, but what I tried involved draping a processor unit around my neck, which was tethered to the ceiling and then to a nearby PC. The prototype headset's a little front-heavy, has a tunnel-vision-narrow field of view, and exposed circuitry everywhere. (Which I was happy about: I spotted at least four cameras, a laser, and what looked like ultrasonic range finders, in case that helps you speculate how it works.)

Sounds like it will be many years before it will be finished. But it will be a key part of Windows 10, the third platform developers should be targeting, alongside PC and phone/tablet.