Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[PRiME]

Yes it would be interesting to have a look at the data for that elephant on UD demo. Now they claim to have some very interesting compression techniques but one would assume such a object with a million points or so would suck up a massive amount of memory compared to conventional models. However if we think about procedural techniques a bit you can see how it could be compressed considerably without loosing much/any quality.

What is the accepted max model file size (including textures) in the industry today for high detail models? 50MB?

[rubbish]

^^ not compression.  He says in the video that the clever bit is the actually the search algorithm that figures out what to display for each pixel.  So you don't necessarily need to use RAM or video card memory to load the models into, the algorithm just pulls the data it needs on the fly so a reasonably fast hard drive is probably more than enough.

[PRiME]

Yes and if the data set is 100GB then its going to be a problem.

[cameni]

Pulling data from disc for rendering? Are you aware of disc transfer latency and speed? There's a reason drives have RAM cache on them, you know.

Clever search algorithms like that are used everywhere, OT doesn't load stuff it doesn't need to render the view. But still the most important part is the cache manager that controls and limits the amount of memory, otherwise it would skyrocket and not be able to run effectively at all.
He may have the best algorithm there is (though it's likely just a common octree search cause that's the fastest), but there's no way around the amount of data in the scene.
Unless you believe in Unlimited Compressibility of data as well.

What Euclideon uses to avoid the problem is instancing. The scene is comprised of a limited number of chunks that are being reused all over, some 10x while other smaller features are instanced 100-1000x, and the octree indexes to those chunks instead of containing them. This 10-1000x instancing is where you can achieve high apparent compression ratio. But, of course, the problem is that the stuff repeats all over. Nevertheless, he could be smart and still procedurally assemble a nice scene out of these building blocks, not unlike Minecraft.

What he didn't show yet is that he can rotate these blocks, which is quite easily doable, though it increases the non-instanced memory and thus shrinking the scene again.
But then what? You could rotate pieces of detailed point-cloud ground, but then you have to stick them together to make variable geometry.
There's a reason why their demo video features a flat world with everything aligned the same way. There are so many issues with this that need to be solved, yet he talks mostly about the search algorithm, which is probably the most trivial thing there.

[foxfiles]

 

Now I shut my mouth, at least in this thread.
       Please don't, just keep the discussion constructive...

Dear angrypig, again, where am I destructive? 
I am not much techno-focussed (however a bit involved in CAD & reverse-engineering, so having to work with accurate scanned surfaces, I just know what are large points clouds, decimation, triangles, reconstruction, B-splines and so),
I am more business-focussed... I run and own small businesses/companies, some very " brick and mortar" compared to the subjects we are talking about here.

So, I've just seen this thread about this company, its claims, and saw Cameni's discussion on another forum, all the buzz around this Australia guys...
As I am a curious little business man, apart of the very techno aspects, any company who is having 9 employees and who do not seem to sell anything since many years,
is an interesting question to me : that's all.
Now as there are many smart and open-minded guys here on your forum, I thought that it was a good place to chat.

Coming back to the techno part, an intricate 3D statue model, scanned from the real object, with good detail and texture, then triangulated from the points,
is at minimum 20MB in Obj. + Png. , even if I am not a specialist, I am, me too, more than doubtful of what they say they are capable of.

[angrypig]

Dear angrypig, again, where am I destructive? 

It wasn't meant to you personally, it was meant for believers who don't need to read and just believe that magic is possible. I don't want to stop discussion here but i prefer one that is constructive and with people that aren't afraid to read papers and asking questions (like you) and do not blindly follow some guy who just told them exactly what they want to hear...

In my post I also wrote that i see possibility to use voxels for some parts because as i said before they have a few nice properties that can be very useful. But there is also the other side of the coin and i prefer if the developer is not hiding a weak side of the technology. Perfect example is Brano Siles with Atomontage engine (AE) who openly says there are weaknesses and also says that he will try to solve them in future to achieve render quality for FPS views etc...

AE latest dev blog post:
http://www.atomontage.com/?id=dev_blog&pid=dec26_2011

[foxfiles]

Don't worry, there is no problem : Yes it is sure that on this forum, there are various kind of people, from several horizons, and it is also very positive that you're attracting so many guys : it is a proof of that many like a lot what you are creating & developing. I saw your super technical approaches on other threads too 
Thank you for the link, I had already seen the impressive Atomontage's works, but I will read more again about them.