Proceedings of the 2007 symposium on Interactive 3D graphics and games
Seattle, Washington
Pages: 73 - 80
Year of Publication: 2007
ISBN:978-1-59593-628-8
(ACM Digital Library)
http://portal.acm.org.www.lib.ncsu.edu:2048/citation.cfm?id=1230100.1230113&coll=ACM&dl=ACM&CFID=45639009&CFTOKEN=87624794
(ACM Digital Library)
http://portal.acm.org.www.lib.ncsu.edu:2048/citation.cfm?id=1230100.1230113&coll=ACM&dl=ACM&CFID=45639009&CFTOKEN=87624794
a) what did you find interesting or novel about the paper?
ReplyDeleteI found the whole paper novel, since I've never known how ambient occlusion works, much less how to do it efficiently. Ambient occlusion is a nice effect, and adds a lot of depth and density to objects.
b) what aspects of the paper were most difficult to understand?
What is a full-screen quad? Other than that, it wasn't very hard to understand after a few readings.
c) was the paper well written?
Yes, I found the diagrams helpful and the descriptions of the equations and methods fairly easy to understand.
d) could the methods have been improved?
Of course, but it's hard to recommend anything, not being a graphics expert.
The occlusion in figure 6 looks very unnatural.
Figure 10 seems like an unfair comparison. It seems to be comparing a completely unshadowed image to an occluded image, when (I think?) the point of occlusion is that normal shadowing isn't enough.
I wonder if it would be possible to determine whether two normal vectors intersect, so to speak, and perform more occlusion on vertices whose normals intersect more? Illustrated here: http://zipt.org/images/normal_occlusion.png
e) what possible applications does this have?
In circumstances where speed is important (always!), this method can perform ambient occlusion faster than other methods (according to the paper). Ambient occlusion is excellent at giving objects a realistic feeling of density.
--
mwc
(Do anyone else's arrow keys not work in this textarea?)
(My arrow keys don't work either, nor can I copy/paste to/from it... Also it hides extra spaces. Bizarre.)
ReplyDeleteI found the paper to be moderately interesting. I recognize ambient occlusion to be something I have seen a good amount of, but haven't noticed. (I suppose it's one of those things that one only notices if it's failing.) Thus I definitely appreciate their performance-oriented approach; it does not seem reasonable to expect readers to go and try out a computationally intensive approach for such a subtle effect that can be easily acheived with raytracing.
I understood the intuition behind ambient occlusion, as well as their choice of spheres to approximate occluders, but I found all the symbols in their math a bit difficult to keep straight in casual reading. The diagrams were okay (I guess) but figure 4 seemed like a waste of space (what's it trying to portray, anyway?)
They never showed their techniques used in conjunction with any other type of shadowing. That makes me wonder how well their technique looks with other shadows. Also, the shadows in their results seemed a litle exaggerated, but their darkness is probably easy to tweak.
They kept using the phrase "visually pleasing" but made no attempt to elaborate on the characteristics of visually pleasing ambient occlusion. Not all of their results were visually pleasing to me - the trees and colosseum on the last page looked good (though the comparision to shadowless images was laughable) but the classical statue in the upper left-hand corner of the first page had some fairly severe outlining around her feet.
I like that they focused on GPU based rendering. I'm sure that this was not necessarily a novel concept, but that they were one of the earlier studies in the frontier.
ReplyDeleteThe funny thing is that throughout the paper I was asking myself - so what exactly is this occlusion thing. Glad wikipedia exist.
I wish they would have pointed out the artifacts more, I finally noticed in figure 1c at the very top that the shadow for the car is really only being generated using the one edge of the car as there is light coming through the wheel well.
I'm not that big into equations so - I disliked those parts of the paper. I also felt the diagrams coul have been explained a lot better - in the figure with the man the box and two spheres - are these spheres part of the scene or is something being covered by them?
This moethod for realtime raytracing seems very feasible on todays graphics cards. It would be great to see a game engine attempt to implement this.
(btw - I just tried posting in IE to see if I still had issues with arrows and spaces and its not present in IE)
It is interesting how the authors have focused on an algorithm that implements high-frequency and low-frequency occlusion on order to implement near and far illumination effects. They use spheres to approximate and avoid surface discontinuity. I do not consider that their algorithm is very novel, they certainly focus on getting the maximum from the GPU hardware, but the results are not very convincing to me.
ReplyDeleteThe paper was in general well written because they separate their ideas in a really organized way. However I think that it was not easy to understand since they left many things unexplained, for example,all the algorithms that they cire in order to give a solution for the problems that their method has in terms of efficency and proper occlusion are not explained at all.
I am not an expert in this area and I am not familiar with the cited algorithms, but by the way they mentioned it, the Depth-peeling algorithm's efficency is very unclear to me. I consider that the ND-buffer is an interesting approach, however I feel like it leaves many unsolved features, for example, when many occluders exist for a pixel, then not necessarily the closest one to the camera is represented.
Finally, ambient occlusion is a general and fast approximation to global illumination; therefore it has a lot of applications in the video game and movie industries. Despite their method was not very convincing to me, I consider that the low-frequency approach brings a fair solution for far illumination, in this way this specific method can be a efficient approach that can be used in parts where the objects are not very close to the camera.
This article addressed the issue of ambient occlusion. It's
ReplyDeletefascinating how programmers can manipulate lighting
effects to make objects appear more realistic. The idea presented,
however, doesn't seem novel, but rather an enhancement of
older methods.
The lay out of the article was nice in that there were
detailed pictures and equations. The authors also formatted
the paper so that it was easy to follow.
Ambient occlusion could be applied to movies and
video games to make objects appear more realistic.
The idea that this algorithm is compatible with most
graphics hardware makes this algorithm flexible as
well.
I found the idea of using the GPU to accelerate ambient occlusion to be novel, but at the same time, because their technique relies on the GPU to be powerful, it isn't as groundbreaking as some of the other papers we read this week. If they had thought of this idea three years earlier, it wouldn't have worked because graphics cards were not powerful enough to render these scenes at more than a couple frames per second.
ReplyDeleteWhen the authors go into detail about their techniques, it becomes a little hard to follow, but over all the paper reads pretty well. They had good use of images and captions to illustrate their points.
I don't know enough about ambien occlusion to say if they could have improved their methods.
Applications that would benefit from this method are real time renderers and games.