For some reason I'm thinking this paper is related to Microsoft's Flight Simulator game. It seems utterly amazing that they managed to compress the entire US geologic dataset into under 400MB. Also this was done on hardware that is now 4-5 generations old and managed 640x480 @ 60fps. Even though they listed MMORPG's as a suitable candidate for this tech I think it could lead to some artifacting - such as you spinning to fast and they a lower level of detail to keep up with you, but there's an NPC that was standing on a bump that now is smoothed in a bit so it looks like he's floating.
I wish they had some shots of close detail, but I'm sure it is relatively good as it's generated from real data sets.
This article addresses the issue of rendering terrain height-fields in real-time. The novel components of the geometry clipmap are decompression in real-time and synthesis in real-time.
The overall layout of the article was nice and easy to follow. Although I had some difficulty understanding the concepts, the majority of the paper was easy to understand. From looking at the results of using the geometry clipmap, the level of detail on the terrain rendered seems to greatly increase.
As said in the article, this application can help improve the quality of video games (more specifically, flight and driving simulators) or movies with outdoor scenes.
This was a very comprehensive paper; the authors covered a lot of material, from generating and updating clipmaps to compressing, texturing and culling them, using them to synthesize detail, and at the end benchmarking their method against other techniques. They did a good job making a full presentation of lots of material without losing too much focus on their main topic.
At first I wondered why they wouldn't cut the corners on their regions, making them octagonal instead (out in the corners there's a lot of needless detail). But they answered it pretty well by explaining their toroidal array and cropping operation - it wouldn't work well with nonrectangular regions.
Flight simulation and geographic visualization were about the only real applications I could think of. Modern games that have large terrains like this already implement background loading, but maybe there are enough advantages to this method (i.e. much less disk access, easily integrated texture LOD) that a developer would consider using it.
The authors give what I think is a very novel and useful method for real-time synthesis and rendering of terrains. It is really interesting how they implement their method extracting the information from a mipmap, that stores the information on a different way I had never seen. They choose nxn grid called the clip region from the mipmap that directly depends on the viewing position since is centered on it. Their method is able to shift the "active region" or clip region in order to match the perspective given by motion/
In general, the paper was well written. They do not focus on giving many mathematical formulas but they give a detailed explanation of their steps in the method, which I find very helpful. However, one of the most diffivult parts to understand is how the mipmap information is extracted.
The algorithm is based on object density for rendering, therefore it is expected that an object that has small surface area/volume is rendered late or inaccurately.
Lastly, this technique can be applied to simulators, video games, films and any field related with cartography or topology.
a) what did you find interesting or novel about the paper?
The dynamic compression is pretty outstanding.
b) what aspects of the paper were most difficult to understand?
All of the concepts were fairly straightforward, I thought. The new concepts for me didn't pose much of a challenge (such as clipmaps, frustum culling,
c) was the paper well written?
It was a little dense, but mostly clear.
d) could the methods have been improved?
It wasn't clear to me if the authors were using frustum culling only to save on rendering time, or if they were also using it to avoid unnecessary decompression. If it's on the former, perhaps the culling could be applied to eliminate decompression of non-visible texture.
e) what possible applications does this have?
In the words of the authors: "real-time outdoor games include flight simulators, driving simulators, and massively multiplayer games". It's possible that some artists would also be interested.
I was astonished to read how the authors are able to compress a height map of the entire United States into 400 MB with only 1.8 meters of error. By putting the entire map into memory, they can ensure that the GPU pipeline is always filled. This allows them to render maps at good framerates.
I found the geometry clipmap and the rendering algorithm to be hard to understand. The material on terrain synthesis was easier to follow because they had good pictures to illustrate their points.
The paper was well written. It specifically mentioned previous approaches to the problem and explained their contributions to the field. It also lists the advantages that their method has over the others.
According to the authors, certain types of terrain are difficult to render in realtime, such as very tall structures. However, this is hardly seen in natural terrain, so it is not an issue.
The most obvious application of this work is in flight simulators. It would be great to fly across the entire continental US without any map loading of drop in framerates. In fact, they may be using this isn Microsoft Flight Simulator already, because I believe that is possible. Another application would be for space exploration. Satellites could take detailed maps of the Moon and Mars and then they could be used in simulators to help train astronauts in landing their landing modules safely.
For some reason I'm thinking this paper is related to Microsoft's Flight Simulator game. It seems utterly amazing that they managed to compress the entire US geologic dataset into under 400MB. Also this was done on hardware that is now 4-5 generations old and managed 640x480 @ 60fps. Even though they listed MMORPG's as a suitable candidate for this tech I think it could lead to some artifacting - such as you spinning to fast and they a lower level of detail to keep up with you, but there's an NPC that was standing on a bump that now is smoothed in a bit so it looks like he's floating.
ReplyDeleteI wish they had some shots of close detail, but I'm sure it is relatively good as it's generated from real data sets.
This article addresses the issue of rendering
ReplyDeleteterrain height-fields in real-time. The novel components
of the geometry clipmap are decompression in
real-time and synthesis in real-time.
The overall layout of the article was nice and easy
to follow. Although I had some difficulty understanding
the concepts, the majority of the paper was easy to understand.
From looking at the results of using the geometry clipmap,
the level of detail on the terrain rendered seems to greatly
increase.
As said in the article, this application
can help improve the quality
of video games (more specifically, flight and
driving simulators) or movies with
outdoor scenes.
This was a very comprehensive paper; the authors covered a lot of material, from generating and updating clipmaps to compressing, texturing and culling them, using them to synthesize detail, and at the end benchmarking their method against other techniques. They did a good job making a full presentation of lots of material without losing too much focus on their main topic.
ReplyDeleteAt first I wondered why they wouldn't cut the corners on their regions, making them octagonal instead (out in the corners there's a lot of needless detail). But they answered it pretty well by explaining their toroidal array and cropping operation - it wouldn't work well with nonrectangular regions.
Flight simulation and geographic visualization were about the only real applications I could think of. Modern games that have large terrains like this already implement background loading, but maybe there are enough advantages to this method (i.e. much less disk access, easily integrated texture LOD) that a developer would consider using it.
The authors give what I think is a very novel and useful method for real-time synthesis and rendering of terrains. It is really interesting how they implement their method extracting the information from a mipmap, that stores the information on a different way I had never seen. They choose nxn grid called the clip region from the mipmap that directly depends on the viewing position since is centered on it. Their method is able to shift the "active region" or clip region in order to match the perspective given by motion/
ReplyDeleteIn general, the paper was well written. They do not focus on giving many mathematical formulas but they give a detailed explanation of their steps in the method, which I find very helpful. However, one of the most diffivult parts to understand is how the mipmap information is extracted.
The algorithm is based on object density for rendering, therefore it is expected that an object that has small surface area/volume is rendered late or inaccurately.
Lastly, this technique can be applied to simulators, video games, films and any field related with cartography or topology.
a) what did you find interesting or novel about the paper?
ReplyDeleteThe dynamic compression is pretty outstanding.
b) what aspects of the paper were most difficult to understand?
All of the concepts were fairly straightforward, I thought. The new concepts for me didn't pose much of a challenge (such as clipmaps, frustum culling,
c) was the paper well written?
It was a little dense, but mostly clear.
d) could the methods have been improved?
It wasn't clear to me if the authors were using frustum culling only to save on rendering time, or if they were also using it to avoid unnecessary decompression. If it's on the former, perhaps the culling could be applied to eliminate decompression of non-visible texture.
e) what possible applications does this have?
In the words of the authors: "real-time outdoor games include flight simulators, driving simulators, and massively multiplayer games". It's possible that some artists would also be interested.
--
mwc
I was astonished to read how the authors are able to compress a height map of the entire United States into 400 MB with only 1.8 meters of error. By putting the entire map into memory, they can ensure that the GPU pipeline is always filled. This allows them to render maps at good framerates.
ReplyDeleteI found the geometry clipmap and the rendering algorithm to be hard to understand. The material on terrain synthesis was easier to follow because they had good pictures to illustrate their points.
The paper was well written. It specifically mentioned previous approaches to the problem and explained their contributions to the field. It also lists the advantages that their method has over the others.
According to the authors, certain types of terrain are difficult to render in realtime, such as very tall structures. However, this is hardly seen in natural terrain, so it is not an issue.
The most obvious application of this work is in flight simulators. It would be great to fly across the entire continental US without any map loading of drop in framerates. In fact, they may be using this isn Microsoft Flight Simulator already, because I believe that is possible. Another application would be for space exploration. Satellites could take detailed maps of the Moon and Mars and then they could be used in simulators to help train astronauts in landing their landing modules safely.