Slashdot Mirror
Experiment Shows Stylized Rendering Enhances Presence In Immersive AR
An anonymous reader writes William Steptoe, a senior researcher in the Virtual Environments and Computer Graphics group at University College London, published a paper (PDF) detailing experiments dealing with the seamless integration of virtual objects into a real scene. Participants were tested to see if they could correctly identify which objects in the scene were real or virtual. With standard rendering, participants were able to correctly guess 73% of the time. Once a stylized rendering outline was applied, accuracy dropped to 56% (around change) and even further to 38% as the stylized rendering was increased. Less accuracy means users were less able to tell the difference between real and virtual objects. Steptoe says that this blurring of real and virtual can increase 'presence', the feeling of being truly present in another space, in immersive augmented reality applications.
Since we are living withing a simulation already this doesn't surprise me.
"A natural next step would be to add haptic feedback allowing users to touch virtual objects. Users could pick up physical items and computer generated ones at the same time while still thinking both are real. Adding the ability to walk around would expand one’s sense of presence as well. This allows individuals to explore computer generated environments further immersing them into the experience." - Article
Yeah, it's for porn
It isn't terribly surprising that adding a cartoonish rendering effect to both real and virtual objects would make them more difficult to discern as such. I certainly wouldn't call it more immersive - quite the opposite, in fact. It is extremely obvious that what you are looking at has been altered and that you are not looking at "reality".
"The more you distort things, the less you can tell them from fakes." Surprise surprise surpriiiise!
Table-ized A.I.
The summary has misquoted from the video caption. 56% is not "(around change)", it is "(around chance)".
This made me think immediately of the Borderlands games, which uses a black outline shader. Lots of PC gamers turn off the black lines (which are separate from the cell shading, by the way) but I left them on because I liked them, for reasons I couldn't ever explain very well.
From the video, the lag was horrible. I guess I see why the whole VR scene is focusing on the lag issue.
"Boffins have found that when you alter the appearance of an object, humans find it more difficult to perceive it as it actually is".
Your thin skin doesn't make me a troll
Not even the article author knows what's real any more! Quote from TFA: For example, users avoided simulated boxes in one of the experiments when walking around despite knowing that they were real.
It reminded me that I needed to upgrade my video card.
"Dude, the colour depth out there is fucking *amazing*!"
Political debates have me rolling my eyes so much I think I got optical whiplash. I should sue. - Foamy The Squirrel
Making everything look crappy results in people not being able to tell what's real. Mainly because everything looks crappy.
In the next experiment, the test subjects were wearing a blindfold under the VR goggles and the recognition dropped to exactly chance level! The VR presence has been perfected.
Combine this with your 3D printer so you actually think you've 3D printed a car.
Alternate Headline: Experiment shows stylized rendering degrades quality to the point where you can't tell which is degraded reality and what's fake.
If they're serious about research, they should analyze the information that has been removed from the real ones to try to understand what the fakes are missing.
(e.g. Animats mentioned the weird lag that made the fake ones stand out like sore thumbs when the scene rotated.)
Just a guess: Poor simulated material properties and indirect lighting are probably some of the bigger giveaways for the fakes.
accuracy dropped to 56% (around change)
Then I watched the video in the article, where they actually say:
Participants demonstrated 56% accuracy (around chance)
i.e.: 56% is pretty close to the 50% you'd expect from just guessing. That one letter makes a big difference.
A recursive sig
Can impart wisdom and truth
Call proc signature()
How is it suprising that throwing an outline around real objects (thus removing shading and light-edge cues) makes the objects harder to differentiate as real/virtual?
Consider also the source and the publisher. (Yes, I actually RTFP)
So, making a natural image look artificial makes it harder to distinguish from artificial images?
In other news, putting a cat inside a box makes it harder to distinguish that cat from a box, covering McDonald's burgers in ketchup makes them taste less like bad meat and more like bad ketchup, and water was found to be rather wet.
Stylised rendering might make it harder to detect the difference between real and virtual objects but clearly at the (great) expense of realism. What seems very obvious to me is the difference in latency between updating the scene for a real object and updating the scene for a virtual object. The virtual objects seem to be catching up with the real objects when the subject pans their view. That 'lag' combined with our natural sensitivity to motion is the problem.
What is a rag & bone man doing in this field anyway?
The summary is deceptive (what a shock). It implies that if you make the rendering of the virtual objects worse, they feel more "real". That could be evidence for the existence of the "uncanny valley" - you'd naively expect that better rendering makes things more real, so something (the Uncanny Valley) must be interfering. But what the experiment actually did was distort the real objects until users had trouble telling them from the virtual objects. That's not nearly as interesting.
However, a variation of this setup might still be useful for investigating the Valley. Compare heavily distorted real objects with lightly distorted ones - which ones are users more likely to treat as fake?
If you do find an Uncanny Valley effect, it would be interesting to see if it applied more strongly to human faces than to inanimate objects.
TFA didn't make it harder to identify virtual objects as TFS implies - It made real objects look more like virtual ones.
Aka, "if we make everything look like cartoons, people can't tell which cartoons came from the real world".
Is that a fancy phrase for "out of focus"? "Low definition"? Why does this "scientific" study evoke a huge, resonating "DOH!" ??
military HUD designers demonstrated that simple raster lines that showed threats in one shape and friendlies in another, allowed for better absorption of critical details than a photorealistic rendering of all data that radar/FLIR/whatever sensors were returning. All the extra crap, like pushing polygons in the latest BlackMedalofWarfare shooter not only confuses the mind but distracts it.
And after that, you can apply "Uncanny Valley" phenomena. If you get it "close" to "perfect" a flaw drags one out of their state of disbelief or leaves distracting "discomfort" that affects the experience. As long as your environment is consistent and smooth, "stylized" works better because you run your system smooth, and can eliminate the flaws your system can't handle.