Title: optimized multi-pass method that could make 3d raymarching 10x faster? Post by: Mrz00m on March 15, 2012, 11:07:39 AM Hello, thanks for creating this amazing piece of software.
This suggestion may only be relevant for specific formulas although it could also make things faster across the board, it's probably been brought up a long time ago but there isnt a thread on the forums for it so here goes: speckle detection for any 2 individual adjacent pixels, or any radius areas with speckles, followed by a second DE pass on the areas to make a more smooth image. it could save 90% of the processing time for high res pictures? Title: Re: optimized multi-pass method that could make 3d raymarching 10x faster? Post by: hobold on March 15, 2012, 11:39:12 AM Such an algorithm would work part of the time. But at other times, it would get confused in areas that have actual, fractal, detail. So it would still require human intervention (or a much smarter way of recognizing which areas of the image need correction).
Title: Re: optimized multi-pass method that could make 3d raymarching 10x faster? Post by: DarkBeam on March 15, 2012, 12:57:18 PM Such an algorithm would work part of the time. But at other times, it would get confused in areas that have actual, fractal, detail. So it would still require human intervention (or a much smarter way of recognizing which areas of the image need correction). True! :yes: Title: Re: optimized multi-pass method that could make 3d raymarching 10x faster? Post by: fractower on March 15, 2012, 04:43:02 PM I think Mrz00m may be on to something. A second DE pass over problematic regions would either remove speckles if the first surface is well defined, or produce another collection of speckles if the surface is fractal. It seems in the worst case the second pass would just waste time, but he is trying to buy a 10x speed up. Giving up a little efficiency seems like a fair trade off.
Title: Re: optimized multi-pass method that could make 3d raymarching 10x faster? Post by: MindsEye on March 17, 2012, 10:53:50 PM True! :yes: You could select a region or outline the area with the mouse to be "despeckled". Just a thought. Title: Re: optimized multi-pass method that could make 3d raymarching 10x faster? Post by: Jesse on March 17, 2012, 11:07:45 PM Of course a 2nd pass or even more was also in my mind, but suggesting a 10x speed gain is not very serious :dink:
Doing the first pass at 10x speeg gain, you will miss critical parts, that you might be not able to restore in further passes. Title: Re: optimized multi-pass method that could make 3d raymarching 10x faster? Post by: Mrz00m on March 30, 2012, 02:18:05 PM Of course a 2nd pass or even more was also in my mind, but suggesting a 10x speed gain is not very serious :dink: Doing the first pass at 10x speeg gain, you will miss critical parts, that you might be not able to restore in further passes. Ok ok. i was going to say a 5x speed gain but i though 10x sounds more dramatic! yes when there are veils and membranes in the parameter they get totally skipped on a fast render. so it would make sense to have user controls for a base level of speed after which to scan to for missing pixels, i think that 10x could be attained as a maximum speed gain for higher res pics, although the average would be indeed just 2x to 5x faster probably. If a base level of detail for a high res pic can be 1 minute, and then only 5 percent of that pic needs to be recalculated at 10x precision, then it could be 18x faster :D |