Oculus + unlimited detail (Euclideon)

Well
I think this is great for having a fluent and homogeneous experience whatever the details are.
Its perfect for a streaming.

But,

Occulus interest is mainly in the latency, the "input lag" and I dont think having network/cloud architecture is a good thing.
Even if the calculation of data in itself will be fast, the time that the Input of the occulus goes to the Cloud, then that the result is computed, and comeback by the network, will be, in my opinion too long.

So there will be better performance (FPS, details) , but higher input lag and for VR with occulus rift, where
input lag is far more important than details due to the screen resolution problems I'm not sure it would be a good solution.

Interesting post - Yeah I never really buy the 'sudden blanket solution syndrome' :arrow:

I find it interesting to revisit the engine debate with OVR on the scene. Real-time Path-tracing solutions like Brigade are certainly future proof concepts but have a long way to go to usurp rasterization solutions; These will be around for some time still due mainly to an extensive development history and galactic feature lists. A voxel solution seems to be the black sheep here, despite being far more appropriate for dynamic scenes than path tracing which struggles there for obvious reasons. Despite this I find PT most interesting for static scenes where absolute realism and quality are paramount such as in Visualisation or some simulations.

"nukemarine" wrote:
I remember seeing their first video about infinite resolution a while back. I've been impressed but reserved.

The first video of theirs that I saw was back around 2003/2004 when they were trying to get funding from local (brisbane, australia) game developers. Still waiting for a single demo to be released showing it actually working. :)

I saw it running at the InterGeo 2013 so I can confirm that it works (although until then I did not believe the snake oil sales man ;) ) . The guy said they had a 5 TB pointcloud that was streamed from a HD while moving the camera around. This means that "unlimited" stands for the size of your HD.

I think it could be good for looking at 3d laser scanner data, but I guess that it can not handle dynamic 3d pointclouds.

A one-post account, ending with "Oculus + Euclideon = EPIC WIN and the future of simulation technology."

Seems legit.

"Sounds simple enough then. Use Euclidean for the non-animated evironment then put the animated models in that environment."

A couple of points in response to that:

First, the trend in games is towards more dynamic, animated, destructible, interactive environments.

Second, it's evident from the demo videos that the Euclidean tech can do two things: it can 'search' a single monolithic, static point cloud, or it can multiply-reference chunks of point cloud within a 3D grid, a bit like old platform games where levels were built out of repeated bitmap tiles.

The single point cloud, which is the option that lets you actually build an environment freely, is impractical. There's simply too much unique data. Twenty billion points might sound like a lot, but to achieve a tolerable walking-around resolution you would need to dedicate around a million points per square metre of surface. So that's 20,000 square metres of surface - which again sounds like a lot. But that's a flat square less than 150 metres on a side. The more intricate you make a shape, the greater the surface area. Make that flat square out of cobble stones instead and you're probably looking at under 100 metres on a side, or an empty box 40 metres on a side. Or ten empty rooms 12 metres on a side. And that's assuming you just store a single colour per point, with no surface normal or specular info for real time lighting.

The multiply-referenced grid system, which is what they show in their 'game' videos, attempts to get around that by re-referencing the same set of data over and over, like having a hundred shortcuts to the same folder on your desktop. The trouble is, each reference is obliged to be exactly identical to every other reference AND at the exact same orientation in world-space. So although you can get a lot more mileage out of your 20,000 square metres of surface, the repetition is going to be painfully obvious - as indeed it is in their demos. They said the demos looked bad because they are 'not artists', which is like saying you came into work without pants because you're 'not a tailor'. The simple fact is that no artist could make that engine look good, because the limitations are so severe.

What they've done is unquestionably very clever, particularly the fact it streams so quickly from 'slow' media like hard drives or USB sticks. It can definitely do some things much better than polygonal techniques. The problem is that they're things you don't want to do in games.

"zanzamar" wrote:
someone with the skills, MAKE IT HAPPEN please!!! Oculus + Euclideon = EPIC WIN and the future of simulation technology.

If they released an sdk and it was suitable, I'd make it. I've been critical of the Euclideon claims over the last decade, but I'll give it a go, I like alternative rendering systems. However they don't want the public to have it. They've never given out demos and no company has ever made a product with it (that's public at least).
Now that they've jumped markets to geospatial visualization, they probably won't be releasing free sdks. It's more likely they will be assisting companies making proprietary tools.

There's some very interesting tech going on in there, I'd love to experiment with it, but if they never give it out, there's not much we can do.

"kojack" wrote:
... but if they never give it out, there's not much we can do.

Except build a clone with the same (or similar) functionality. I have put a LOT of thought into this (hundreds of hours), and I have privately shared my ideas in detail (in a long series of PMs) how this could be done. Not ready to share openly yet, but I will give a little hint: Think "BWT indexing" to lookup the color of each on-screen pixel... The person(s) with whom I had that discussion became convinced that my ideas would work. Of course, ideas are not what counts, but there actual implementation, and THAT costs hard-earned blood, sweat, and tears (but mostly time, lots and lots of time)...

From a fixed view point, the algorithm would be very simple (there's some youtube videos that show how it could be done). But moving... that's harder. :)

I believe one of my students is (or was) working at Euclideon, I could ask him how it works if I could remember which student (from years ago) it was. :)
(One of my other students mentioned it, but I forgot who he said it was, and now I can't remember who told me either)
Not that I would, I'd never ask somebody to break NDA/privacy/etc. But I still really want to know how they are doing it. :)


Something else that isn't mentioned much, their current system takes hours to process point clouds into a suitable format. It's coming back to Quake style level editing (I remember some quake maps that took up to 24 hours to generate the lighting and visibility data on a typical home pc).

Euclideon currently have a patent called "A METHOD FOR EFFICENT STREAMING OF OCTREE DATA FOR ACCESS", that could be a hint. :)
http://pericles.ipaustralia.gov.au/ols/auspat/applicationDetails.do?applicationNo=2012905545
But it costs money to get the actual patent docs from the australian patent office.

It works not unlike this: [URL Removed by cybereality]
A novelty is the existence proof of the superfluity of rays, /, * or floats in regard to rendering. UD or Atomontage & all mystery mongers alike are going, eventually, to be floored by the above modest beginnings.

VR/AR is evil, will make kids even more exposed to all species of filth, allow NATO soldiers to free flickies out of wicked jihadist robots (violence-free slaughter courtesy of AR) and, as you all know (wish?), mainly used for base purposes.

Everyone gets what they deserve.