I ran across this today, and His use of a mesh to display His example made me think of garagecoder, dinasset, Souphead, and Ronounours .

This looks like a good candidate for The GIMP!, and or G'MIC.

He has already implemented it in Python using PIL (Pillow) so I could use his code as a base and convert it to a plug-in, but I do not understand it, thus would not be able to intelligently make the effect alterable in any way. For example, the vertical offset of the altered edges, or the degree of curl involved. This looks similar to the page curl effect in some existing plugins/scripts, but seems at first glance to be a superior implementation.

http://benpaulthurstonblog.blogspot.com ... ormal.html

If it's based on creating a map of locations to "pull" pixels from then that would translate to G'MIC well (i.e. an absolute warp). My knowledge of matrix math is close to nothing sadly but I'm willing to learn, the question is how long it would take. I'll wait and see if anyone else is going to hop in and have a go, there's little point me taking an unknown amount of time on it if it takes somebody else 30 minutes

As a side note, wasn't there something like this already coded for the next version of gimp? My mind invents thing sometimes so I may have imagined it

I haven't been keeping up on next version items, so I am not sure about that. In fact, I only considered doing a 2.9 install today, and maybe a roll your own as well, still thinking on it.

I have been looking at this for some time now (around an hour) though, and I am beginning to understand it. Slightly. Still not something I could tackle, but I am working on it. Also firing a message to author asking for some help. His blog is full of in depth math articles, so I am assuming a math/computer science background. I just hope He can "dumb it down" enough for me. I'm good at math, but I don't speak the language.

I had a quick look at the actual program and it seems fairly trivial to do a "blind" conversion, I think it's distributing pixels from the source hence the oversampling. Ideally if it could be inverted to calculate for each output point an (x,y) source location there wouldn't be a need for oversampling.

OK, that comment is beyond me as I do not understand the "oversampling" concept. Right now I am stumbling on meanings of the variable c. The explanation led me to believe it is a point, but it has way to many coordinates for a single point ..... Need a Deep Thought here.

c is an array of (x,y) coordinates representing the vertices a to h (in this case hard-coded)
The oversampling is literally going over the points of the source multiple times to avoid holes in the output

Edit: I'm wrestling with the wiremap filter and some bizarre alpha channel behaviour from gimp, so I won't be working on the above any time today!

LOL! No problem. I threw this in here for a consideration thing. I don't actually need (or really even want) this. I just thought others might be interested, so created this thread for input.

As for myself, I just want to understand it. The concept is solid, and it might have other applications. Imagine an animation made of this ...

I wish we can have the "Curve-bend" plug-in working like this.

@jazzon:

Did you ever hear anything back?

Turning this into a "warp" - so it can be interpolated instead of oversampled - has a lot more to do with inverse quadratics than matrices (they're only for convenience). I have a working inverse "E-bezier" warp now but the rest will likely take me months of learning, which I don't mind but it's a long wait.

Perhaps somebody good at maths can help (Souphead/David)?
Failing that I'm tempted to contact the author myself...

garagecoder
Ewwwww... thats ugly lets try this one, at least it isn't ugly!
garagecoderNo I have not goten a response yet.

Purely for demonstration there's an example under "testing -> garagecoder -> PQCT Example" due to appear within about 25 mins. Note there's no oversampling in this version, but still valid enough to show aliasing caused by distributing pixels instead of interpolating, which is my point about needing inversion.

What I now know though is the new G'MIC math parser is amazing
It's fast enough to use for a type of 2D point cloud, thanks to being able to access other images within a fill (see the hacky bit at the bottom of the filter code).

This isn't optimised, literally chopped up the example so it could be speeded up still.

Haven't looked yet but I'm Gonna!!

Garagecoder here are my results with the test.



Results-

Here is the effect results if i keep running the filter on a duplicate.

Thanks for adding an example Rod

You'll notice holes (black pixels) in the image, oversampling would mostly solve that problem but it's not ideal. Still, it could be useful so might as well finish it off - I should be able to make it faster which will offset the penalty from oversampling.

I'm not sure what to do about controls though, sliders for that many points would be unwieldy. Interactive would be best but possibly a lot of effort!

When I run the filter in G'MIC, I get an error.
This is the error I get:
*** Error in ./gcd_pqct/*repeat/*local/ *** Command '-f': [_cimg_math_parser] gmic<float>::fill(): Invalid item 'q(s,t,a,b,c,d,e,f,g,h)' in expression '...,7,1)); q(s,t,a,b,c,d,e,f,g,h)= ((1/4)*b-(1/2)*s*b+(1/2)*s*s*b+(1/2)*t*t*d+(1/2)*t*d+(1/4)*e+(1/4)*g+ t*t*s*s*((1/4)*g+(1/4)*b+(1/4)*d+(1/4)*e)-(1/2)*t*t*s*g+(1/2)*s*s*g+ (1/2)*s*g-s*s*((1/4)*g+(1/4)*b+(1/4)*d+(1/4)*e)-t*t*((1/4)*g+(1/4)*b+ (1/4)*d+(1/4)*e)-(1/2)*t*t*s*s*e+(1/2)*t*s*s*e+(1/4)*t*s*f+(1/4)*t*s*s*f+ (1/4)*t*t*s*f+(1/4)*t*t*s*s*f-(1/4)*t*t*s*c-(1/4)*t*s*s*c+ (1/4)*t*t*s*s*c+(1/4)*t*s*c-(1/2)*t*t*s*s*g-(1/2)*t*e+(1/2)*t*t*e- (1/2)*t*t*s*s*b+(1/2)*t*t*s*b+(1/4)*d-(1/2)*t*s*s*d-(1/2)*t*t*s*s*d+ (1/4)*t*t*s*s*a-(1/4)*t*s*a-(1/4)*t*t*s*a+(1/4)*t*s*s*a-(1/4)*t*s*s*h+ (1/4)*t*t*s*s*h+(1/4)*t*t*s*h-(1/4)*t*s*h)...'.

Just how is the script supposed to work?

@mackenzieh

I think G'MIC 1.6.9 is required, because it uses one of the latest features in G'MIC: functions. This might change after I optimise things, but as the filter stands it's not worth worrying about (there aren't even controls yet)! Really it's just a quick test to try it out.

Oh, okay. I've seen the pre-release for the new version but not the actual release for it. Is it even out yet?

Yes it is out. If You are running a Ubuntu 14.04 or derivative, it requires some tweaking to install though due to new libraries needed.

You could disscuss interactive with David I know G'MIC has the capability of opening a interactive window now.