# technohermit

Member

11

136 Neutral

• Rank
Member
1. ## Sphere - Surface collision elimination problem

Um, if you save the distance between the objects for one frame, then next frame simply test to see if the +/- (sign) of the previous distance and the current distance are opposite. If they are, you DEFINITELY had a collision! If not, you can perform the normal test (collision = )distance < ball-radius)). :-)

3. ## real 3D-geometry vs fake (normal/relief/parallax/POM=parallax-occlusion-mapping/etc)

Yes, maybe, probably. But unless somebody does a serious and careful analysis, we are just repeating the conventional wisdom that is being reasonably questioned here. Any honest person *must* admit, it is easier to repeat the conventional wisdom and say it is "obviously" correct, than to carefully and methodically "prove it" - or make an actual case for it. I do not think ANY of us has done that yet, though the cobblestones and scales and similar cases are clearly the best place to start. So far, we are just hand-waving and repeating conventional wisdom. Admittedly, that is oh-so-much easier, but are we really certain that 3D geometry cobblestones and scales are *substantially* worse to render as 3D geometry? I ask it this way because we are ALL certain that 3D geometry is the best way to display an awful lot of things. If we display everything that way, we could put all our efforts into optimizing and polishing that code --- and guess what, far less switching shaders back and forth between so many different shading techniques. Which is much better from the proven "fewer batches is better" point-of-view.
4. ## real 3D-geometry vs fake (normal/relief/parallax/POM=parallax-occlusion-mapping/etc)

Here is my two cents to the replies so far. The original post admits the presented example was chosen to "stress the point" slightly, so POM comes out looking really horrible in many ways. However, is it actually clear that POM and POM-like techniques are ever substantially better in fully modern, up-to-date 3D graphics hardware. Because, if we find it is difficult to find cases where POM is clearly significantly "better" (faster/cheaper/whatever), the point of the original post becomes "due to changes/advances in the speed/memory/shaders in next generation 3D graphics cards, 3D geometry is rarely significantly slower or inferior to the so-called faked-geometry techniques". So, is it truly clear that POM is worth doing in the context of currently best graphics-cards - which will be mainstream in 1~2 years and old-hat in 3 years. Here are the cases where POM seems like it *might be* better (to some degree). Where almost all of the geometry is near the highest level - like in many cobblestone walls or sidewalks. In these cases, the camera-view-ray will usually intersect the geometry within two or three iterations - even at the slow pace of 2~3 pixels per step. In a case like this, you still benefit from POM over normal-mapping, because there are deep grooves between cobblestones which can exhibit substantial parallax that normal-mapping cannot represent. I had a couple other situations in mind, but now I realize full geometry does work better in those cases. Still, the above case does look to me like a possible win for POM --- as long as none of its inherent defects matter. Which means, the camera must not fly down into (or nearly into) those grooves between the cobblestones, because this becomes more like the originally described case where 3D geometry wins. And the camera should never get down close to the cobblestones, because then the "streaky" character of the near-vertical grooves in the cobblestones become obvious artifacts of fakery. So, I guess we must carefully examine this case that seems to be a win for POM given the context and limitations I mention above. If we cannot prove this truly is a win for POM (or relief/parallax/similar-mapping), then the original post seems pretty much entirely valid. Fact is, quite surprisingly, the old-fashion way *does* look alot better than might be expected, I think, and for exactly the reasons he suggested, probably. In thinking about his idea, I realized it seems entirely practical to make a very convenient LOD scheme based on the geometry approach, based upon what is pretty much the conventional efficient approach today with OpenGL IBO/VBO pairs (or DX index-buffer/vertex-buffer pairs). What you do is this. You have all the vertices in one VBO/vertex-buffer, plus one IBO/index-buffer for every LOD you want to support. Each IBO references just those vertices in the one VBO that it needs to construct the object at the desired resolution. Presto, you have LOD with only one set of vertices. This is surely not original, but hey, it works. So, where are we?