Slashdot Mirror
Euclideon Teases Photorealistic Voxel-Based Game Engine
MojoKid writes Not many would argue that current console and PC graphics technologies still haven't reached a level of "photo-realism." However, a company by the name of Euclideon is claiming to be preparing to deliver that holy grail based on laser scanning and voxel engine-based technologies. The company has put together a six-minute video clip of its new engine, and its genuinely impressive. There's a supposed-to-be-impressive unveil around the two minute mark where the announcer declares he's showing us computer-generated graphics rather than a digital photo — something you'll probably have figured out long before that point. Euclideon's proprietary design purportedly uses a laser scanner to create a point cloud model of a real-world area. That area can then be translated into a voxel renderer and drawn by a standard GPU. Supposedly this can be done so efficiently and with such speed that there's no need for conventional load screens or enormous amounts of texture memory but rather by simply streaming data off conventional hard drives. Previously, critiques have pointed to animation as one area where the company's technique might struggle. Given the ongoing lack of a demonstrated solution for animation, it's fair to assume this would-be game-changer has some challenges still to solve. That said, some of the renderings are impressive.
It looks really good, it's fantastic actually.
However, I could still easily tell that these were not real world images. Some more than others stood out, but in motion these elements still looked 'wrong' for real.
I think his work is fantastic and state of the art, but I think he was a little to hopeful in his video that no one could tell until he said it.
The thing about voxel raytracing is that it usually requires less fancy programming and design work to get your graphics up to snuff.
Rasterization, while extremely efficient, requires layers upon layers of programming cleverness and hours of skillful modeling and texture creation to pull off a "photorealistic" look(let's be honest, it's not that good either). If you could just throw out all the lightmapping and the real-time-self shadowing hacks, and the ambient light simulation, and a bunch of other stuff that's cropped up over the years to make up for the fact that we're not raytracing, you might choose to.
Indeed! Lighting has always been their biggest drawback. Their static lighting, even using lightmaps, is a FAR cry away from Dynamic Global Illumination.
I'm much more impressed with Unreal Engine 4's realtime global illumination and and architecture real-time
Another thing the Eucliedian guys don't get is is that polygons are "good enough." Photorealism is a red herring of computer games. All the photorealism doesn't make a game FUN, only the potential to NOT break immersion. When game design breaks immersion the easiest all the photorealism in the world won't save it -- ironically, it makes it worst! What do you mean I can't climb that fence?
Cool tech of course with their compression for voxel data, but until they have real time dynamic lighting and global illumination ... *yawn*.
This guy has an annoying, self satisfied way of speaking that just makes me want to beat the snot out of him.
Voxel graphics are interesting and the laser measurement plus automatic texturing from a real world scene is cool, but this just does not compare in detail or framerate to a mesh generated by the exact same laser scanner and a little bit of pre-processing, all of which has been possible for over a decade now.
Plus, what are you going to do with this 3D scene? An interactive game? But games need dynamic objects, which cannot really be done well with voxels and will contrast dramatically with the scene's lighting model. You don't have any light probes, or spherical harmonic coefficients or anything useful for static lighting dynamic objects, let along dynamically lighting static objects.
When Argumentum ad Hominem falls short, try Argumentum ad Matrem
I think the stuff at the end is fairly close to photorealistic, assuming the photo is low-resolution. What it *isn't* is camera-realistic. Their camera engine uses pans and zooms that in no way reflect how a physical camera would move through real space -- this makes the entire effect look fake.
What they needed to do to make the demo a "wow" demo is put the camera inside the physics engine, and give it the mass and movement of a real camera. The results would have been much better.
The one bit of the video where I thought "hey, that's actually decent" was on the zoom-in on the stair tread, as the zoom was similar to what you'd get on a camera on a tripod, and the stair looked pretty photorealistic. For the rest, our brains enter the "uncanny valley" not because of the images presented, but because of the surreal way they are presented.