hi forum~
this is what i want to do:

1. load an input (pix_film/pix_movie/pix_video = ok).
2. remove colors / make parts of the input transparent (pix_alpha = ok).
3. apply this modified input to a rectangle geo (pix_texture = ok).
4. then *extrude* this 2d texture into a simple 3d geo (like a cuboid) while preserving the alpha channel (i.e. transparent parts of the image look like 'holes' in the 3d object).

i am really not sure how to proceed with 4. i stole the image below as the closest example i could google (although it doesn't demonstrate the alpha preservation).

my first thought was to use pix_texture on a cuboid, however this doesn't seem to be the solution as it applies the 2d texture to all 6 surfaces (i.e. i guess a cuboid has no 'front' or 'back'). i want the 2d texture to stay facing the camera and extrude the texture back farther into the 3d space. or is there a way of modifying the default pix_texture behaviour on 3d geos? like having control on how it fills the 'Z' dimension...

is there an OpenGL object that achieves this effect?
glBindTexture?
should i install an OpenGL external? http://linas.org/gle/index.html
should i switch to blender or something?

help!

you probably already know/use this patch but [pix_coordinate] looks like it'd come in handy here. i never quite understood the whole deal with the s/t params, but it makes for a fun kaleidescope effect (http://puredata.hurleur.com/sujet-661-gem-kaleidescope-patch)

i'm pretty sure i get what you're trying to do but figuring out how to do it is gonna be tricky i bet - i'm not sure what patch achieves the "transparent" or "opaque" effect you're looking for, and then on top of that you need to figure out a way to stretch it out to a 3d model/figure and retain the transparent qualities. if you can figure out how to implement that openGL external you linked to then maybe it would be possible to just have two copies of the same pix_texture (be it a still or a video or a webcam) wrapped onto two 2d squares, have one on top of the other with a decent gap inbetween them, and then figure out a way to "fill" the areas between the 'active' colors on the Y axis - wrapping the texture to a cuboid wraps it to each side, and besides that i don't think an object like cuboid would be flexible enough for this.

basically what i'm thinking is take the image you posted and instead of trying to deal with one 3d block (cuboid) maybe use 2d squares, 2 of them with the same texture, and place them like the top and the bottom of the cuboid (like the bottom and top of a box just floating in space). since these two squares contain the same image figure out what object you need to fill the 3d space between the 'active' colors (according to how you set your color threshold, that extern looks like it might be handy for this). have it fill along the Y axis between the active pixels in the image, and then you should have basically the effect you're looking for. a benefit to this method (if it is even a feasible one, i'm just thinking out loud at this point) is that you would really even need a 'transparent' effect - the transparent bits would just be spots not filled because of their color threshold not triggering the fill. then you could take the 2d squares that act as the limits of the fill and just make them invisible one way or another - that way you could spin it all around in 3d and you should be able to see 'through' it from every angle!

all in all i would say it's definitely possible, it's just a matter of what method would work best... i can see there being multiple methods of achieving the desired effect, but you also have to make sure to not be redundant and waste processing overhead (like, say you used a transparency effect on a 2d square and then [repeat]ed it a few hundred times). i just threw that idea out there because i'm not sure that you can treat the default 3d objects to transparency like that...

hope that made sense, and i also hope i understood what you're trying to do correctly (otherwise that's one long useless post i just made ... if i'm on the right track let me know and i'd be willing to help you out with it. either way i just had a lot of fun thinking that out!

@slvmchn said:

basically what i'm thinking is take the image you posted and instead of trying to deal with one 3d block (cuboid) maybe use 2d squares, 2 of them with the same texture, and place them like the top and the bottom of the cuboid (like the bottom and top of a box just floating in space). since these two squares contain the same image figure out what object you need to fill the 3d space between the 'active' colors (according to how you set your color threshold, that extern looks like it might be handy for this). have it fill along the Y axis between the active pixels in the image, and then you should have basically the effect you're looking for.

hmm maybe that hack could work. i will post if i have success...

i also thought about using LOTS of [pix_data] objects analyzing the color of the texture, and drawing geos from that (in effect, reconstituting the video input with added Z axis). but as [pix_data] can only analyze one pixel at a time it seems tiresome and too much pointless work (even for pd surely there's a better way...

what about PDP? the 3dp_push function? i can't test on linux for a few days.
OpenGL brainiacs?

@halftheory said:

i also thought about using LOTS of [pix_data] objects analyzing the color of the texture, and drawing geos from that (in effect, reconstituting the video input with added Z axis). but as [pix_data] can only analyze one pixel at a time it seems tiresome and too much pointless work (even for pd surely there's a better way...

[pix_data] can only analyze one pixel at a time... Well, any operation proceeds 'one at a time,' I suggest you add recursion to your gem render chains. I've done something similar to what halftheory proposes in this manner:

1. Load a video.
2. Using [t a b] and [repeat 16] Build a basic counter that increments from 0-15. The repeat object is to make sure that your counter counts to 15 asap, EVERY FRAME.
3. Now, translate the 0-15 into a matrix which will scan your video, something that [pix_data] will understand.
4. Dump the pixel data into arrays.
3. Build a separate gem chain for your polygons. Again, using recursion and counters to create a 4x4 (16 total objects) 'grid' (2d, or 3d) of polygons.
4. Use a counter to read from the arrays you created in step 4, to set the color, texture, translation, rotation, whatever of your grid.

This is a quick and dirty of my common process for utilizing video as an 'organic' (analog sourced) control structure for manipulations of recursive polygons. If you are interested, I could go into depth.

Found an example in the bin. This patch scans a pix as described above and then applies the RGB values to a 4*4*3 grid of cubes.

http://www.pdpatchrepo.info/hurleur/rgbGrid2.pd

Hey Nestor, thanx fro the patch - it helped alot

To all: I am trying to do pixelation of a live-video (actually the luminosity would be enough in the first place) and I've adapted the patch to suit my needs (s. attachment). But - being a stupid artist an no programmer - my grid logic is floored.
The goal is to give a neural net a simplified image (made out of 48 fields in this case) to train gesture recognition. Has anyone got an idea on how to do this with pix_data?

thanx alot for any help

http://www.pdpatchrepo.info/hurleur/rgbGrid23.pd

im afraid i dont really get what you want to do...i have a patch which does GLSL displacement to a meshsquare or obj (wich you can texturize as you wish in an 3d app) based upon a greyscale texture, like a movie or live video.

so what it actually does is push out the verticals of the (XY-oriented meshsquere) in z direction.
i renderd some z depth images in 3ds and it looks cool. and i can send if this is want you want.

a cuboid only consists of 8 verticals, so displacement will only affect these 8 verticals, for a bump map like thingy you will need some mesh with more subdivisions, so make a appropriate obj in blender, 3ds or wherever.
thanks to Iohannes gem supports uvw coordinates.

to get nice transparency, use tiffs with alpha channel as texture and put an *alpha* in the chain.

ok, now i got it...cubes! that patch is cool.

as long as you dont need tons of cubes ( like 256x256) pix_data will be ok.
but its realy slow. if you want more, glsl is the way to go.

you can see the patch i was talking about before here

Yeah, great - I have seen this on Processing but not yet in PD - very nice. But what I'm looking for is something like this: http://h3.gd/ann/perceptron.png - in the long run I want to connect every "Mega-Pixel" to 1 neural input. But in the first place, I'd really love to simply "pixelate" the video down to 8 by 6 Pixel from pix_video live-in for re-projection on dancers during a show.
But yours is great to, love to understand how you did it. Btw: Are you German? There is now a pd group on facebook (do not like FB but anyway)... nice to chat faster.

Thnx 4 fast replay.
Enkidu