Botond Kósa
|
|
« Reply #195 on: June 04, 2014, 01:34:50 PM » |
|
Not true, you can trivially do it in smaller sub-blocks or tiles, buckets, whatever you want to call them. This is only true for IFS fractals, where you cannot simply ask for the colour of a (sub-)pixel.
In fact, because of this convenient property for escape-time fractals, you don't even need to store the full 1:1 image in memory, just some small buffer so that you aren't saving the pixels to disk one at a time (which you can compute in file-format order and save directly).
Of course you can render the image in smaller blocks. What I wrote applies to the current implementation of MM only.
|
|
|
Logged
|
|
|
|
|
SeryZone
Strange Attractor
Posts: 253
Contemplate...
|
|
« Reply #197 on: June 09, 2014, 11:00:09 PM » |
|
Hello, Botond! How to realize Marianni/Silver algorithm for simpliest z^2+C??? I tried to read post on the sites, but do not understand. Can you help me?
|
|
|
Logged
|
|
|
|
Botond Kósa
|
|
« Reply #198 on: June 13, 2014, 11:48:42 AM » |
|
Thanks, that location with the glitch in one corner proved that calculating SA coefficients based on image width is not always enough, since the corners are farther away from the center than the edges. Using the diagonal instead of width produces a correct image.
|
|
|
Logged
|
|
|
|
Botond Kósa
|
|
« Reply #199 on: June 13, 2014, 12:24:36 PM » |
|
Hello, Botond! How to realize Marianni/Silver algorithm for simpliest z^2+C??? I tried to read post on the sites, but do not understand. Can you help me?
In the classic Mariani/Silver algorithm you calculate the pixels along the border, and if all have equal iteration values, the interior can be filled with that value. (This can be done because of the connectedness of the Mandelbrot set and the dwell bands.) If not all pixels are equal, divide the region into smaller pieces (dividing in half along the longer edge is best), calculate the iterations on the divider line, and apply the border equality check for the subregions recursively. If you are calculating normalized iteration counts (a.k.a. smooth gradients), this method can only speed up the interior of the set, since all outer pixels get different iteration values. But it can be tweaked to work on fractional values as well: instead of checking the equality along the border of a region, check for monotonity along the four edges. If the (fractional) iteration counts grow steadily along all 4 edges, the inner pixels' iteration values can be interpolated. I tried several interpolation functions (e.g. bilinear) and measured the mean absolute error to find the best one. Keep in mind though that the Mariani/Silver algorithm is only effective on relatively sparse images with large monotonic areas. These areas appear on low zoom depths or on images with a narrow iteration span where series approximation is very effective. On hard images with complex, dense structures and wide iteration span, Mariani/Silver doesn't help much. You can observe the effects of the algorithm by checking the Hide guessed pixels checkbox under Rendering in Mandel Machine.
|
|
|
Logged
|
|
|
|
Roquen
Iterator
Posts: 180
|
|
« Reply #200 on: June 24, 2014, 10:47:04 AM » |
|
TL;DR.
FYI: On my old 64-bit notebook it goes boom in mm64.dll executing an illegal instruction (AVX probably...didn't check).
|
|
|
Logged
|
|
|
|
Chillheimer
|
|
« Reply #201 on: July 16, 2014, 04:04:17 PM » |
|
any idea what is happening here? am i doing something wrong?
(attached a zipped mmf-file of location)
|
|
|
Logged
|
--- Fractals - add some Chaos to your life and put the world in order. ---
|
|
|
3dickulus
|
|
« Reply #202 on: July 16, 2014, 06:47:42 PM » |
|
It looks the same as when my GPU tries rendering beyond precision limits.
|
|
|
Logged
|
|
|
|
|
youhn
Fractal Molossus
Posts: 696
Shapes only exists in our heads.
|
|
« Reply #204 on: August 05, 2014, 08:22:48 PM » |
|
any idea what is happening here? am i doing something wrong?
This looks exactly the same as what happened here with Kalles Fraktaler just a few days ago. It wasn't even a very deep zoom. After zooming out and in again, the problem was still there. After killing the program and restarting from the same location, the problem was gone and zooming continued as normal. Since the same location did well after restart, the cause is not in the location. So I did not keep it. For Mandel Machine .. I wonder how the development is progressing. Botond, could you provide us with a short status update?
|
|
|
Logged
|
|
|
|
Chillheimer
|
|
« Reply #205 on: August 05, 2014, 08:31:23 PM » |
|
It looks the same as when my GPU tries rendering beyond precision limits.
seems like precision limits have something to do with it, when I set this from auto to a manual value I could zoom deeper (but encountered other problems)
|
|
|
Logged
|
--- Fractals - add some Chaos to your life and put the world in order. ---
|
|
|
stardust4ever
Fractal Bachius
Posts: 513
|
|
« Reply #206 on: August 06, 2014, 12:05:00 AM » |
|
Is there some way to extend the precision limit of Mandel Machine. I posted the following render today: http://www.fractalforums.com/images-showcase-%28rate-my-fractal%29/pterodactyl-vertebrae/The current feature is located at 3610 zoom levels. Mandel Machine appears to have a hard limit of 3700 zooms. This is part of a render path I plan on continuing a point I estimate to be just under 5400 or so zooms. Mandel Machine is so fast and slick it's literally unreal but I need absolutely need to go deeper. I'm currently carving my way through using Kalles Fraktaler 2 but if there's another software app comparable to Mandel Machine, I'd like to use that instead. Kalles Fraktaler 2 has a few glaring issues that make it clunky to render finished images. For starters, the aspect ratio seems to be pegged at 16x9 and there doesn't seem to be a way to rotate the image to fit onscreen, especially if it's an oblong shape which as Murphy's law dictates, that the shape wil be vertically aligned instead or horizontal.
|
|
|
Logged
|
|
|
|
Botond Kósa
|
|
« Reply #207 on: August 06, 2014, 12:09:21 AM » |
|
I have to say, your gallery on deviantart drove me to improve the speed of Mandel Machine. Before implementing the perturbation algorithm, it was already 30-40% faster than Fractal Extreme at zoom depths beyond 500. (This can be tested by unchecking "Use perturbation algorithm" under Computation.) It is truly unbelievable you had the patience to find such deep locations like Magnum Opus Ex, using the conventional algorithm with full precision calculations.
|
|
|
Logged
|
|
|
|
stardust4ever
Fractal Bachius
Posts: 513
|
|
« Reply #208 on: August 06, 2014, 12:43:59 AM » |
|
I have to say, your gallery on deviantart drove me to improve the speed of Mandel Machine. Before implementing the perturbation algorithm, it was already 30-40% faster than Fractal Extreme at zoom depths beyond 500. (This can be tested by unchecking "Use perturbation algorithm" under Computation.)
It is truly unbelievable you had the patience to find such deep locations like Magnum Opus Ex, using the conventional algorithm with full precision calculations.
What do you think got me burned out of doing fractals, LOL! I spent over a month rendering that "Magnum Opus Ex" image at 3200x3200 pixels! My latest render (see post above) is deeper (3610 zooms) and took like 32 seconds to render at 4096x3072 using Mandel Machine, even on the same CPU which is now 2.5 years old - like, days literally became seconds - how is this even possible !!! Is there any chance you could increase the zoom limit? Pretty please...
|
|
« Last Edit: August 06, 2014, 01:53:25 AM by stardust4ever »
|
Logged
|
|
|
|
Botond Kósa
|
|
« Reply #209 on: August 06, 2014, 12:57:28 PM » |
|
... For Mandel Machine .. I wonder how the development is progressing. Botond, could you provide us with a short status update?
The recent developments were focused around 3+1 key areas: 1. Automation of calculation parameters (90% complete): many controls in the Computation box are there to provide manual override in case the image gets distorted. This happens mostly around minibrots. There are also some optimizations possible under certain circumstances that provide some speedup. When finished, users will no longer have to fiddle with computation controls, and the GUI will become simpler. (The hidden controls will still be available in some "Advanced" mode.) 2. Speeding up the initialization of series approximation (100% complete): the computed iteration count can be largely reduced by applying more aggressive series approximation (see Reply #165 in the thread). But the time needed to initialize the approximation is a quadratic function of the number of terms, so the right balance has to be found. By speeding up the initialization by up to 5 times, more aggressive approximation can be used. 3. Re-enabling features disabled in 1.2 beta (25% complete): automatic glitch correction and saving of iteration count data for the history items were disabled, which reduced the functionality of the application. They will have to be enabled to come out of beta state. +1: several bugfixes and some GUI improvements I am planning to release another beta after arriving back from holiday (next week). That will include the first two feature sets. Re-enabling the disabled features and releasing a fully functional version will take some time, a few weeks at least.
|
|
|
Logged
|
|
|
|
|