Gridless advection

[JaydenDP 6]

Hi everyone!

Any idea's kicking around to perform good stepped gridless advection at render-time for volume rendering?

Just an open topic for no reason at the moment in particular. Was just thinking about it as at the moment we do use render-time advection on our volumes which does one noise displacement (no history for multiple steps). At least I think it does that, I didn't write it ;P

Any thoughts and musings would be much appreciated anyway.

Thanks

Jay

[Annon 80]

I took one look at gridless advection and shrunk away in fear. Deadline was far too tight to even contemplate it!

Watching this closely...

[SpencerL 28]

In your shader, you could use a For VOP. Inside it have your noise. Each iteration, increase/decrease the timestep of the noise offset (depends if you want forward or backward tracing). Make sure you add to the current location of P so that it creates the feedback.

[Jordan Walsh 10]

I have been playing with some of the ideas mentioned in this thread:

Especially Serge's awesome setup here:

page__st__36#entry84164

Is this considered gridless advection?

[nord3d 9]

This topic interested me since I first saw the Jerry Tessendorf's paper "Resolution Independent Volumes". Recently, I was able to get some results with houdini native resources. I has tested a few different methods, the simplest is just displace volume with velocity field to sharpen the resulted density. Later, I going to tell about this in more detail, now I only post some pictures (the method(s) is still under development).

The simple displace method (before, after):

More advanced method:

[Jordan Walsh 10]

nice one nord3D.

Im very eager to see more!!!

[nord3d 9]

Now, I placed the video of this experiment here:

[ikarus 67]

Interesting, care to elaborate on the process? Are you just evolving the field a number of steps equal to the frame number?

I would guess that the dissipation is coming from the fact the density output isn't coupled with the evolving vel field, so you lose the divergence-free projection. You see the same effect when advecting particles through a field over a long period of time.

[nord3d 9]

Interesting, care to elaborate on the process? Are you just evolving the field a number of steps equal to the frame number?

Yes, exactly.

I would guess that the dissipation is coming from the fact the density output isn't coupled with the evolving vel field, so you lose the divergence-free projection. You see the same effect when advecting particles through a field over a long period of time.

Thank you for the tip, I'll think about it.

Another pictures, from another algorithm, not so good for now, but (I hope) can be improved...

Edited December 4, 2012 by nord3d

[nord3d 9]

My experimentation continues, so I have a few new pictures:

#1 without / with

#2 without / with

#3 some frames from the sequence in analytical color

The speed is greatly improved and the algorithm is greatly simplified, but still has some problems. I'll working on it.

[whodini 15]

Is there any chance you could perhaps share some of the approaches or hip file for the algorithm you are using in the above images?

[johnLIC 5]

I too am quite curious. Nice stuff!

[Jordan Walsh 10]

So Im still wondering what the definition of gridless advection is for someone who cant quite decipher the white papers yet... anyone got a good description in laymen's terms?

[whodini 15]

wow, it's been over a year. Can anyone shed some more light on this topic? the tessendorf papers are pretty tough to digest. Has anyone made any attempts to make sense of it and willing to share?

[jkunz07 190]

Advect density field by velocity field in a for loop in cvex context (volume procedural).

 

Here is some psuedo code for the rough idea:

 

for i in $F

@density += volumesamplev("/path/to/velocity/field_" i ".bgeo", "vel", @P)*@TimeInc;

 

 

So you use the "i" variable to read in specific frames of your vel field

Doing this in cvex at rendertime, density can be moved around resolution independant or "gridless advection"

Edited April 10, 2014 by jkunz07

[Jordan Walsh 10]

So this you are kind of calculating the advection for every frame leading up to the frame you are rendering... that sounds pretty heavy computationally... 

[eetu 526]

Ok, I haven't done this, but I don't think you can "push" density around without storage. You would need to "walk backwards" in time, through each frames (negated) velocity field, and see if you end up somewhere that has density.. RIght?

[ikarus 67]

Ok, I haven't done this, but I don't think you can "push" density around without storage. You would need to "walk backwards" in time, through each frames (negated) velocity field, and see if you end up somewhere that has density.. RIght?

 

yeah basically you do a 'trace' advection setup, walking backwards in time to find where the fluid would be in the next step.

 

This example's pretty basic, I would want to add some modifiers based on density and velocity magnitude to make the step size more adaptive.

 

 

rendertime_advection.hip

[teratera 9]

Thanks, ikarus! It works really well. Render time seem reasonable too, for this image it was 0:08 / 0:22 at 8 steps, of course low-res, so probably not the best example to test. Technique is very cool, thanks again, so much abilities there.

[whodini 15]

Thank you ikarus for generously sharing a hip file! that's incredible. I owe you a beer man. This is extremely valuable for learning volume shading and more.

Edited April 14, 2014 by rhussain