Jump to content

Euclideon Interviews and new video (Holoverse, Unlimited Detail tech, Games...)

Go to solution Solved by mschuetz,
On 26.9.2016 at 1:47 PM, EUD said:

I asked the dev again, hope he answers how the potree works. 

I am the dev of potree.

 

Edit:

Ok, I found your question in a  youtube comment in this video (not from me) that I missed.

 

Streaming in potree simply means to send chunks of points with different levels of detail over the web to the client, in order to gradually increase the details. Each chunk usually contains around 1,000 to 10,000 points. UD certainly does the same thing only with voxels, and that is why you see the progressively increasing details, especially with slow connections. 

Streaming from a remote source to local RAM always sends chunks of points, voxels or pixels. Sending just a single element with each packet would be way too inefficient since each packet that is being sent over the web has a significant overhead associated to it. Same if you stream data from a local disk to RAM. It's not feasible to load single points from disk. Many of the loaded voxels or points in a chunk may be invisible right now but that's okay because they are likely to become visible when moving or looking around so it's good to have them in RAM anyway. 

 

Edit2: 

Quote

 and the streaming of chunks is not even done the way other renderers do it and is only present IF THERE IS LATENCY FROM INTERNET OR SOURCE OR THERE IS TOO BIG LOAD ON SOURCE.

Again speculating but the reason this streaming process is fast in UD videos is because they're compressing the data and because they won't be using some slow remote servers for their videos. For most models in potree you need to load around 3 million points for any given point of view to get a pretty good level of detail. Since there is no compression right now, each point requires 16 byte so we have to load 50megabyte in total. If you're streaming directly from disk and not over the web, you'll load the data in less then a second. Once you got the initial data loaded, you'll have to load less data while moving around and a little more data when rotating. That's because when you move forward a little, you've already loaded most of the data and you only need to load a little more for your new view. You can even do prediction and fetch the data for what you believe will be visible in a second or two, based on the direction of movement. 

 

Now 50mb for 3 million points is only because there is no compression in potree. It's relatively trivial to get this down to 25mb and euclideon claims to use realy good compression so maybe they got it down to something in the range of 10mb. Besides, sparse voxel octrees compress better than point clouds at lower levels of detail (SVO: low precision at low resolution, point cloud: full precision even at low resolution), so it may or may not be slightly less than that. 10mb for the initial view is very quick to load and after that it's only some additional data as you move around. 

It is, however, not instant. Rotating by 180 degrees is close to a new view and while doing so you will be able to notice the streaming progress. Same if you move fast. Even UD won't be able to stream in the data fast enough during quick movements, especially not if they want to achieve a level of detail close to current AAA games. 

5 hours ago, Pohernori said:

 

Enlighten us. In text please.

I am. What do you want to know.

Link to comment
Share on other sites

Link to post
Share on other sites

11 minutes ago, EUD said:

Data, the points being streamed. Potree is not streamed, i talked to dev on youtube. Its just intended for web viewing. Its cool, useful, nice but not UD

What do you define "streaming" then? You're not making much sense.

 

Talking to dev and saying something is intended for web viewing =! potree is not streamed.

 

I load a potree demo, cut off internet, the points stops loading, how is that not data being streamed?

 

11 minutes ago, EUD said:

Ray tracing is too taxing.

Yes and voxel based rendering is next to useless in games. At least with the pace API and GPU are improving, ray tracing based rendering will be viable in 3-4 years.

 

11 minutes ago, EUD said:

lol what bro, i think you read somewhere that they have prebaked lighting, and now you are confusing everything. xD

Show me anything, a demo, a video, anything, that shows euclideon rendering (not a scan of a RL environment) with modern game grade lighting, or even dynamic lighting.

Link to comment
Share on other sites

Link to post
Share on other sites

19 minutes ago, crystal6tak said:

What do you define "streaming" then? You're not making much sense.

 

Talking to dev and saying something is intended for web viewing =! potree is not streamed.

 

I load a potree demo, cut off internet, the points stops loading, how is that not data being streamed?

 

Yes and voxel based rendering is next to useless in games. At least with the pace API and GPU are improving, ray tracing based rendering will be viable in 3-4 years.

 

Show me anything, a demo, a video, anything, that shows euclideon rendering (not a scan of a RL environment) with modern game grade lighting, or even dynamic lighting.

fine, if you dont believe me go ask the guy yourself. the points are not streamed from their hard drive to your PC. 

 

i would give you proof, but i have deleted many of my old yt accounts so they are not visible anymore. nor does it states on their page anywhere or youtube that they are streaming points. its only you who makes stuff up. i notice this alot with people biased to hating on Euclideon 

 

modern grade lighting is getting worked on. takes time. UE4 wasnt made in a year you know. takes even more time is zou have a new waz of rendering models

 

and its not voxels

Link to comment
Share on other sites

Link to post
Share on other sites

18 minutes ago, EUD said:

fine, if you dont believe me go ask the guy yourself. the points are not streamed from their hard drive to your PC. 

 

i would give you proof, but i have deleted many of my old yt accounts so they are not visible anymore. nor does it states on their page anywhere or youtube that they are streaming points. its only you who makes stuff up. i notice this alot with people biased to hating on Euclideon 

 

modern grade lighting is getting worked on. takes time. UE4 wasnt made in a year you know. takes even more time is zou have a new waz of rendering models

 

and its not voxels

Oh man, you are a stubborn one. 

 

What is there to believe? Like literally the potree demo stops loading point cloud data if the internet connection is cut off. Get it through your thick skull.

 

Just to humor you, did a quick google search. In a presentation on potree, in a slide listing challenges "Data streaming to the browser"

 

Also, the euclideon engine has been in the works for at least 13 years. Unreal Engine 4 was also in development since 2003. Guess what? You can download UE4 right now and make a game with it.

Link to comment
Share on other sites

Link to post
Share on other sites

I don't understand what problem the point cloud system is trying to solve for games. The need to have "unlimited detail" zooming in isn't really a necessary feature for games  unless you like looking at objects up close most of the time. The only other thing this could resolve is destructibility of objects and models, but at the moment it's only useful in local instances. It's hard to do it for online instances as Battlefield will attest (there are cases where debris from a blown up structure obstruct the view of one player but not the other because the other may have joined the game after the destruction and the debris wasn't counted for).

 

Also there's another thing that Euclideon doesn't tell me, how does it handle lighting? I'm probably wrong, but I get the impression that geometry work is a small fraction of the total time spent rendering a scene. The bulk of the work is due to lighting. Granted "unlimited detail" could skip things like the normal and parallax map lighting steps, but there are still dozens of other lighting steps to take into account.

 

On an aside, I'm finding it hard to believe this claim:

Quote

a few million draws per frame

So this man figured out how to make a few million draw calls on OGL and maintain "high FPS"? Or is "draws" just another word for "I rendered a triangle!"?

 

Also the other thing I want to know is how this affects creating assets.

 

If you really want to replace the current system with this, you need to explain every thing in the design process at how this will either be: A. better and B. save time/money. So far all I see is nothing more than "here's a thing that could really change the world if you tried." If Euclideon really wants to sell their tech, make something that everyone else can poke around with first. Linux didn't get its start by Linus going "I'm making this OS kernel, but I'm not going to show you the source code or anything else other than 'here's a thing'".

Link to comment
Share on other sites

Link to post
Share on other sites

16 hours ago, EUD said:

I am. What do you want to know.

 

The basics that many people don't. And not just repeating what is said in the video.

i5 2400 | ASUS RTX 4090 TUF OC | Seasonic 1200W Prime Gold | WD Green 120gb | WD Blue 1tb | some ram | a random case

 

Link to comment
Share on other sites

Link to post
Share on other sites

On 14/09/2016 at 7:49 AM, Bhav said:

Also note that Euclideon are deleting any comments simply pointing out that their video is showing streamed pre rendered footage and that they still dont have any real time demos.

 

Why would they do that if they havnt been lying?

 

The people falling for the hype on their video comments generally admit to not knowing anything about hardware, thus it isnt that hard to impress people who dont know the difference between streamed crap and hardware accelerated.

Maybe because they aren't lying, and those comments are unfounded, and shutting out actual discussion and attention of the product?

 

I have no idea if their technology is everything that's claimed (I hope it is), but if it really is, and the last time they made a public presentation overwhelming negative presumptions made any progress in public reception of it impossible, wouldn't they be pro-active in preventing the same thing from happening a second time?

 

Why would they allow their work to be drowned out by hecklers a second time, if they haven't been lying?

You own the software that you purchase - Understanding software licenses and EULAs

 

"We’ll know our disinformation program is complete when everything the american public believes is false" - William Casey, CIA Director 1981-1987

Link to comment
Share on other sites

Link to post
Share on other sites

Just to clear things up, potree does stream points. Points are organized in an octree and each node of the octree contains a subsample of the whole data set, with a low resolution subsample at the root and with each level, the resolution doubles. Kind of like an image pyramid. During rendering, additional nodes, which contain roughly 1,000-10,000 points on average, are streamed in to increase details. 

 

I am 100% certain that euclideon does something similar with voxels. There are some videos where, if you move around or rotate, you can see how details are progressively refined from large blocks to very small blocks. You can see the same behaviour in the web demo but even more pronounced than in the videos. This happens because new chunks of data are streamed in; first low resolution chunks and then gradually higher resolution chunks. Might be they're using the infamous sparse voxel octree and each chunk of data they're streaming contains the next 2, 3 or more levels of detail. 

 

In my opinion, euclideon geoverse is a quite nice point cloud/voxel viewer but it's not even close to what they're promising. I've seen it on conferences and was impressed and underwhelmed at the same time. Impressed because it realy is an incredibly fast voxel viewer, underwhelmed because the quality sucked (aliasing, huge problem with point clouds/voxels in general) and because the streaming in of additional chunks was very noticeable which is a no go for games. 

 

So what is point cloud and voxel rendering good for if not for games? To display vast amounts of data generated by laser scanning and photogrammetry. 
Country-scale laser scans will return tens to hundreds of billions of points and surveyers need a way to look at this data. They need to be able to navigate around at idealy 60fps but it's okay if they have to wait a second or two until the highest level of detail is loaded. 

Still looking forward to see high-resolution voxel games and I'm sure the time will come, but I'm not laying my hopes on Euclideon. They're overpromising and underselling. 
 

Link to comment
Share on other sites

Link to post
Share on other sites

On 23. 9. 2016 at 10:10 AM, mschuetz said:

Just to clear things up, potree does stream points. Points are organized in an octree and each node of the octree contains a subsample of the whole data set, with a low resolution subsample at the root and with each level, the resolution doubles. Kind of like an image pyramid. During rendering, additional nodes, which contain roughly 1,000-10,000 points on average, are streamed in to increase details. 

 

I am 100% certain that euclideon does something similar with voxels. There are some videos where, if you move around or rotate, you can see how details are progressively refined from large blocks to very small blocks. You can see the same behaviour in the web demo but even more pronounced than in the videos. This happens because new chunks of data are streamed in; first low resolution chunks and then gradually higher resolution chunks. Might be they're using the infamous sparse voxel octree and each chunk of data they're streaming contains the next 2, 3 or more levels of detail. 

 

In my opinion, euclideon geoverse is a quite nice point cloud/voxel viewer but it's not even close to what they're promising. I've seen it on conferences and was impressed and underwhelmed at the same time. Impressed because it realy is an incredibly fast voxel viewer, underwhelmed because the quality sucked (aliasing, huge problem with point clouds/voxels in general) and because the streaming in of additional chunks was very noticeable which is a no go for games. 

 

So what is point cloud and voxel rendering good for if not for games? To display vast amounts of data generated by laser scanning and photogrammetry. 
Country-scale laser scans will return tens to hundreds of billions of points and surveyers need a way to look at this data. They need to be able to navigate around at idealy 60fps but it's okay if they have to wait a second or two until the highest level of detail is loaded. 

Still looking forward to see high-resolution voxel games and I'm sure the time will come, but I'm not laying my hopes on Euclideon. They're overpromising and underselling. 
 

Again you are confusing stuff. Comon guys its not so hard to understand and they even adress this in explanation that are very simple.
You are here saying that ". Impressed because it realy is an incredibly fast voxel viewer, underwhelmed because the quality sucked (aliasing, huge problem with point clouds/voxels in general) and because the streaming in of additional chunks was very noticeable which is a no go for games. " yet its not voxel based renderer, and the streaming of chunks is not even done the way other renderers do it and is only present IF THERE IS LATENCY FROM INTERNET OR SOURCE OR THERE IS TOO BIG LOAD ON SOURCE. Its not inherent to their algorithm, and if you didnt judge their algorithm by the UDweb demo, you would see there is no chuk-ing going on when you have point cloud stored on your harddrive.

 

 I am not 100 percent sure, but streaming points is different than streaming big packets of points and i am pretty sure potree is not doing former in some way you are describing or whatever. Thats more like downloading than streaming. 

Udweb streams points from hard drive from australia-i suppose to your PC realtime, for each pixel on screen it takes point from point cloud and sends to you, basically directly on your monitor.

 

And euclideons UD is much more powerful, even if potree is cool and runs well. If we all agree potree is cool, how cool would be if it had more power and be able to run games? Hell yeah, wed be all excited. 

I am, for euclideon.

 

Now I might be horribly mistaken right now, but since we know how the UD works, UDweb demo should, theoretically work the same way plus the distance it needs to travel on internet wires.

 

I asked the dev again, hope he answers how the potree works. 

Link to comment
Share on other sites

Link to post
Share on other sites

Ok....as far as i am concerned "Unlimited detail"....i call BS. What is even unlimited?....let me go philosophical a bit here. We are limited beings, computers are limited devices. We cannot contain infinity so there is NO unlimited. Big enough that we cannot tell the difference, that we hit a limit and it doesn't matter if there is more beyond that.....ok, YES, that i agree.

 

Now that we cleared that out of the way. As far as i can tell they somehow found a method of structuring the data really smart so that they can offer - if we can trust what they are saying- sequential access (some form of clustering and connecting the clusters). I really can't imagine how. But anyway, they structure the point clouds is such a way that a searching algorithm can then come and find very very fast the points that correspond to each pixel of the screen. And again if we can trust what they are saying, this happens regardless of how many points there are in the point cloud data. And this is the part where i really get skeptical because that would mean they have a searching algorithm that on the right data runs in constant time (aka O(1) if i'm not wrong).

So u say u have a searching algorithm that runs at the same speed no matter how much data it has to dig through....be it even specially structured for this...

I'm no expert on this but....as far as i know when you manage to do this it will be a BIIG breakthrough in computer science.

So.... i doubt the guys at Euclideon managed to do such a thing. But yeah, they might found a way to come close enough to that point where it doesn't really make a difference, and even if it's not perfect it's good enough that when you look a it (at the results) it seems perfect.

I hope they managed such a thing and innovate. If they did it or are close to....keep up the good work guys. But you'll need to give us more details, more empirical proofs :D until we truly believe you.

 

P.S.

Just thinking now.......all this has the potential to become what No Man's Sky came to be. Cool thing, hype hype hype, people don't actually understand what it will be and have higher expectations than they should and bang...all the disappointment and rage and...well.....just the Internet in general -_-

In the end everything is gonna be ok;

If it's not ok, it's not the end.

Link to comment
Share on other sites

Link to post
Share on other sites

6 hours ago, Nano0806 said:

snip

Unlimited Detail (point cloud), ... is limited... by your monitor resolution. Oh and hard drive space, obviously.

 

This engine already proved itself working well. They just need to develop good game SDK, and from then professional game makers will be able to create... well much better games than Euclideon. Thats not to say there games and minigames are rubbish, its just they are made from stick and stones. Literally. Scanned random objects, put in game, bam environment.

 

If you are interested in, supposedly unlimited detail point cloud renderer similar or even same power than euclideon, check David Eriksson Point cloud simplification and processing for path-planning and Approximate distance queries for path-planning in massive point clouds.

 

 Your first point about Euclideons unlimited claim. Well David, smart guy is doctor of philosophy applied mathemathics. I cite from his CV https://people.cam.cornell.edu/~dme65/cv.pdf" Applied Researcher Fraunhofer – Chalmers Centre, Gothenburg, Sweden Point Cloud Visualization I worked on visualization of huge point clouds with billions of points. It is of great interest to be able to visualize massive point clouds, but limitations in both RAM and graphic card memory make it impossible to render all the points in the point cloud with enough frames per second. The visualization software I developed in C++ is capable of rendering 50+ FPS with unlimited detail on a standard graphics card for point clouds having billions of points. The software reads points directly from the hard disk drive to the GPU when needed."

Link to comment
Share on other sites

Link to post
Share on other sites

Yes i understand. I read most of David Erikssons thesis. I am not saying this can't be done. I like the ideea and i really wish they succeed and bring innovation in graphics processing. What i'm saying is....there is nothing unlimited and in don't think things are all that ideal as it is stated. See my last post...it's kind of complicatedly written but it shows what some people might come expect based on Euclideon statements.

Bottom line....i'm for Euclideon and i wish them all the best and will be looking closely for news on their work. I don't know how they do it, and i know will not be perfect, the be all end all in graphics processing. But they have very big potential and even if they somehow fail miserably(hope not) they will bring in attention a powerful alternative way of doing graphics :)

In the end everything is gonna be ok;

If it's not ok, it's not the end.

Link to comment
Share on other sites

Link to post
Share on other sites

On 26.9.2016 at 1:47 PM, EUD said:

I asked the dev again, hope he answers how the potree works. 

I am the dev of potree.

 

Edit:

Ok, I found your question in a  youtube comment in this video (not from me) that I missed.

 

Streaming in potree simply means to send chunks of points with different levels of detail over the web to the client, in order to gradually increase the details. Each chunk usually contains around 1,000 to 10,000 points. UD certainly does the same thing only with voxels, and that is why you see the progressively increasing details, especially with slow connections. 

Streaming from a remote source to local RAM always sends chunks of points, voxels or pixels. Sending just a single element with each packet would be way too inefficient since each packet that is being sent over the web has a significant overhead associated to it. Same if you stream data from a local disk to RAM. It's not feasible to load single points from disk. Many of the loaded voxels or points in a chunk may be invisible right now but that's okay because they are likely to become visible when moving or looking around so it's good to have them in RAM anyway. 

 

Edit2: 

Quote

 and the streaming of chunks is not even done the way other renderers do it and is only present IF THERE IS LATENCY FROM INTERNET OR SOURCE OR THERE IS TOO BIG LOAD ON SOURCE.

Again speculating but the reason this streaming process is fast in UD videos is because they're compressing the data and because they won't be using some slow remote servers for their videos. For most models in potree you need to load around 3 million points for any given point of view to get a pretty good level of detail. Since there is no compression right now, each point requires 16 byte so we have to load 50megabyte in total. If you're streaming directly from disk and not over the web, you'll load the data in less then a second. Once you got the initial data loaded, you'll have to load less data while moving around and a little more data when rotating. That's because when you move forward a little, you've already loaded most of the data and you only need to load a little more for your new view. You can even do prediction and fetch the data for what you believe will be visible in a second or two, based on the direction of movement. 

 

Now 50mb for 3 million points is only because there is no compression in potree. It's relatively trivial to get this down to 25mb and euclideon claims to use realy good compression so maybe they got it down to something in the range of 10mb. Besides, sparse voxel octrees compress better than point clouds at lower levels of detail (SVO: low precision at low resolution, point cloud: full precision even at low resolution), so it may or may not be slightly less than that. 10mb for the initial view is very quick to load and after that it's only some additional data as you move around. 

It is, however, not instant. Rotating by 180 degrees is close to a new view and while doing so you will be able to notice the streaming progress. Same if you move fast. Even UD won't be able to stream in the data fast enough during quick movements, especially not if they want to achieve a level of detail close to current AAA games. 

Link to comment
Share on other sites

Link to post
Share on other sites

5 hours ago, mschuetz said:

snip

thanks for clarification

 

yea, i didnt take latency into consideration when talking about UDweb, i am so biased >P . 

 

still i might dig a little deeper. 

 

 

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×