[mikiex]

Originally Posted by steampoweredgod

The arbitrary image is turned into precise specific corresponding fractal formula which grants resolution independent representation. Video fractal compression is also possible for arbitrary videos.

It won't add anything that ain't their implicitly(detail that any artist could predict is originally there). The formula is not also infinite it merely has self-reference and is quite finite and simple easily computable.

And would allow the image to remain perfectly legible and understandable, with proper proportions, even if use a giant layer of atoms the size of the distance between earth and the sun to generate a high resolution image bigger than this planet.

That is the nature of fractal representations that you can use more atoms then in the visible universe to generate a screen of x pixels, and the image will be perfectly scaled and legible, as happens with vector graphics representations of text.



dealing with incommesurability, pi the square root of two, and allowing 1 to 1 mapping between diameter or side and circumference or diagonal is beautiful and shows that a digital procedure may generate perfect triangle or circle.
There are 2 documentaries one more ancient the other more novel. Fractal architecture abounds in the solutions of evolution throughout the ecosystem, and even your own human body and brain

Vector text has infinite resolution and it has nothing to do with fractals, its just math

[Simon F]

Originally Posted by mikiex

Vector text has infinite resolution and it has nothing to do with fractals, its just math

Related to this, in a previous life, I helped develop a vector graphics rendering/painting system where the geometry and rendering was done in float (with coverage-based AA). The demos would usually contain models at multiple scales and so things would keep on appearing as you zoomed further and further in....until you got to the point where the floats hit the denorm limit and suddenly all the smooth spline edges became saw teeth!

[mikiex]

Originally Posted by Simon F

Related to this, in a previous life, I helped develop a vector graphics rendering/painting system where the geometry and rendering was done in float (with coverage-based AA). The demos would usually contain models at multiple scales and so things would keep on appearing as you zoomed further and further in....until you got to the point where the floats hit the denorm limit and suddenly all the smooth spline edges became saw teeth!

You found the Atoms then !

I remember in school we had this CAD drawing program on the BBC that had a weird joystick, I was facinated by hiding little bits of text at microscoptic level. Id be quite interested in a game where you zoom in on things.

[hoho]

Originally Posted by steampoweredgod

The arbitrary image is turned into precise specific corresponding fractal formula which grants resolution independent representation. Video fractal compression is also possible for arbitrary videos.

Exactly the same way as you can have hash collisions you WILL have "fractal collisions" when trying to represent something with less data than in the original.

Also,
http://www.femtosoft.biz/fractals/fractal.html
"I've been able to achieve compression on the order of a gif or jpg file."

I tried his delphi app to compress a 1680x1050 minecraft screenshot and it tells me it takes around 1000 minutes on a 2.8GHz i7

Doesn't sound all that efficient to me. Do you happen to have any better stuff I can play around with or does it all exist just in theory?

[edit]
This just keeps on getting better and better.

I tried compressing a 400x220 picture (same screenshot resized) and it gave me 61.1kb file. Original .bmp was 257kb, png 102kb and jpg at 95% quality 39kb so I thought it wasn't quite a disaster at reducing the file size.

And then I decompressed the fractal to find out that it's only grayscale! Same image in grayscale BMP: 86.9kb, png: 48.1kb and jpg 33.9kb. So yeah, a lossless PNG is better than fractal, at least for this image:


It would be fun to test it with actual photos or at least some 2kx2k textures used in games. I wouldn't be surprised if the result is even worse due to minecraft having so huge texel sizes on screen.

[mikiex]

Originally Posted by hoho

Exactly the same way as you can have hash collisions you WILL have "fractal collisions" when trying to represent something with less data than in the original.

Also,
http://www.femtosoft.biz/fractals/fractal.html
"I've been able to achieve compression on the order of a gif or jpg file."

I tried his delphi app to compress a 1680x1050 minecraft screenshot and it tells me it takes around 1000 minutes on a 2.8GHz i7

Doesn't sound all that efficient to me. Do you happen to have any better stuff I can play around with or does it all exist just in theory?

[edit]
This just keeps on getting better and better.

I tried compressing a 400x220 picture (same screenshot resized) and it gave me 61.1kb file. Original .bmp was 257kb, png 102kb and jpg at 95% quality 39kb so I thought it wasn't quite a disaster at reducing the file size.

And then I decompressed the fractal to find out that it's only grayscale! Same image in grayscale BMP: 86.9kb, png: 48.1kb and jpg 33.9kb. So yeah, a lossless PNG is better than fractal, at least for this image:


It would be fun to test it with actual photos or at least some 2kx2k textures used in games. I wouldn't be surprised if the result is even worse due to minecraft having so huge texel sizes on screen.

Take a look at this page
http://users.senet.com.au/~mjbone/Fractals.html

They have links to the early iterated systems exe;s who originally came up with the method and format FIF,
It was combined as a compression / resize originally. But these days the same tech continued "Genuine Fractals" and the emphasis is on the resizing it became "Perfect Resize": http://www.ononesoftware.com

How good is it? Thats a debate, I can't say I've seen a decent real test - normally they just compare it with the some what dated photoshop resampling

[hoho]

Originally Posted by mikiex

Take a look at this page
http://users.senet.com.au/~mjbone/Fractals.html

Their app doesn't work in 64bit windows. I'll see if I can get it to work in VM.

[Shifty Geezer]

Originally Posted by hoho

Doesn't sound all that efficient to me. Do you happen to have any better stuff I can play around with or does it all exist just in theory?

I'm afraid steampoweredgod is not available to answer at the moment. If you'd like to leave your name and number, he'll never get back to you...

Originally Posted by mikiex

How good is it? Thats a debate, I can't say I've seen a decent real test - normally they just compare it with the some what dated photoshop resampling

Even then, the fractal enlargements are clearly upscaled. They aren't filling in any convincing details. I'm wondering if a simple bicubic upscale with a little blur and peturbation wouldn't give the same sorts of results at a lot less effort.

Fundamentally, I'm not seeing any application of fractals in image resizing or storage. They'll remain used for procedural content creation and that's about it.

[Davros]

Steamy's problem is his argument is based on what he imagines fractal compression to be, not what it actually is.

[Gerry]

Originally Posted by Davros

Steamy's problem is his argument is based on what he imagines fractal compression to be, not what it actually is.

I'm not sure that was his only problem to be honest.

[Dr Evil]

Originally Posted by Gerry

I'm not sure that was his only problem to be honest.

What do you mean? Are you saying this isn't normal!?

http://steampoweredgod.blogspot.com/...celerated.html

So it seems people have accelerated their rate of completion of celestial sword project
Then it is time for me to force life through my decaying veins, and start my ritual of ascension, coding, godlike, beyond godlike speed coding, faster than a god of code, faster than the god of the wired, faster than a superintelligence, due to perfect player code base library access hyperturing capability through hilbert's dream, wonderland through the looking glass alice lies.

Dunno seems pretty legit stuff to me...

[mikiex]

Originally Posted by Dr Evil

What do you mean? Are you saying this isn't normal!?

http://steampoweredgod.blogspot.com/...celerated.html



Dunno seems pretty legit stuff to me...

Yar, Interesting Blog..

[ebola]

fractal's place is in procedural textures/geometry
microtextures aren't so far off
see allegorithmic middleware
fractals are ok for filling in the gaps but don't represent true information

regarding the OP you'd just smooth and/or apply jitter to geometry (displacement maps) for resolution independence

[jameszhao]

I'm just pulling stuff out my ass here... but...

So you can project spatial data to a frequency domain with Fourier transforms. Can you similarly project spatial data to some, oh let's call it 'recursive domain' where the basis are recursive patterns? Does it even make any sense?

[ebola]

Originally Posted by jameszhao

I'm just pulling stuff out my ass here... but...

So you can project spatial data to a frequency domain with Fourier transforms. Can you similarly project spatial data to some, oh let's call it 'recursive domain' where the basis are recursive patterns? Does it even make any sense?

micro textures on displacement maps for objects that are supposed to be rough
guess the type of micro texture per texel from the surrounding area of the real modeled detail or allow artists to paint it

[Shifty Geezer]

Doesn't need fractal though. Any suitable procedural algorithm could be used.

[Davros]

A great example of procedural texture generation is .Kreiger
http://www.multimolti.com/blog/2008/...acked-to-95kb/

[Shifty Geezer]

That's not quite true, as I understand it (though I may misunderstand). Isn't Kreiger and others mostly built on the premise of rebuilding a list of artist actions rather than creating content via algorithms? That'd make it a macro-generation system. Things like SpeedTree already use fractals for procedural content. Terrain generation is a classic. There are a few texture creators where you plug in variables and get a procedural texture out the other end, but I don't know of any realtime/in game procedural texture generation. I think fractals will be too processor intensive to have merit for realtime content. Other cheaper noise and pattern algorithms are move serviceable, I think.

[Davros]

your probably right
it doesnt really fit the standard definition of procedural content, but does sort of qualify

[Rodéric]

Perlin noise has been used to generate procedural content in realtime since 2002 or something.

[mikiex]

Originally Posted by Shifty Geezer

That's not quite true, as I understand it (though I may misunderstand). Isn't Kreiger and others mostly built on the premise of rebuilding a list of artist actions rather than creating content via algorithms? That'd make it a macro-generation system. Things like SpeedTree already use fractals for procedural content. Terrain generation is a classic. There are a few texture creators where you plug in variables and get a procedural texture out the other end, but I don't know of any realtime/in game procedural texture generation. I think fractals will be too processor intensive to have merit for realtime content. Other cheaper noise and pattern algorithms are move serviceable, I think.

Unreal used procedual textures

http://wiki.beyondunreal.com/Legacy:...TextureFactory

ignore the word fractal though, it think it was called fire engine

We are going back some yeaaars though

I might try and put in a fractal shader in the next commercial game I'm working on
probably a 1 bit mandlebrot though

[jlippo]

Originally Posted by Rodéric

Perlin noise has been used to generate procedural content in realtime since 2002 or something.

Frontier: first encounter had full planets generated procedurally during runtime in 1995 and it certainly wasn't first game to use procedural generation for meshes. (midwinter 1989 and many others.)

[nbohr1more]

Correct me if I'm wrong but isn't part of the RAGE (Id Tech 5) pipeline oriented around collecting artist actions akin to Kkrieger?

I always thought that was a bit of a paradox. You are collecting artists actions yet baking the end-result to a large unified data-set?

I probably misread that or something got lost in the chain of transmission.

This is really the crux of why Mega-Texture seems to be problematic compared to the standard tiled-texture paradigm. The re-use of texture tiles is "of itself" a form of compression. I wonder if any engines are more amenable to a diversity of very very small textures such that you are essentially using textures like different brushes in a paint app rather than traditional (larger) photo-source approaches.

[mikiex]

Originally Posted by nbohr1more

Correct me if I'm wrong but isn't part of the RAGE (Id Tech 5) pipeline oriented around collecting artist actions akin to Kkrieger?

I always thought that was a bit of a paradox. You are collecting artists actions yet baking the end-result to a large unified data-set?

I probably misread that or something got lost in the chain of transmission.

This is really the crux of why Mega-Texture seems to be problematic compared to the standard tiled-texture paradigm. The re-use of texture tiles is "of itself" a form of compression. I wonder if any engines are more amenable to a diversity of very very small textures such that you are essentially using textures like different brushes in a paint app rather than traditional (larger) photo-source approaches.

I don't know but it could be that authoring the large data set for something like Rage is the problem, I would imaging some system of painting layers - yes you can have totally unique texels, but if you have to cover say 1km and more of textures what are you going to fill it with as an artist? You would want a way to paint it, undo , change and replace textures. If this is true, then you are sort of correct anyway with Rage (if you assume very small textures can be any size you like). But they dont compress it that way (I assume). Combine the fact you might be painting more than one layer, if say you were painting normals/spec/diffuse (Do they do that in Rage?) - you couldnt do each layer by hand, it would take too long.

As far as reuse for compression, that I guess if you think of it, it is kind of what traditional 3d games do, but with a lot of manual work of the artist.

You could of course paint layers record it and then repaint them into a megatexture to decompress them - but the reason not to do that might be that they want the data to be in a compressed format - such as DXT - so it would take too long to do all the work in the background.

[Davros]

Back in the win95 days I used to use an editor called satori (because it came free on a magazine coverdisk and it cost about 4 grand normaly)
anyway its claim to fame was massive image size and unlimited and selective undo because every brushstroke, effect, blur ect was an object in its own right
for example
you could apply Gaussian blur to an image then radial blur then another Gaussian blur and you could undo just the radial blur if you wanted.