H12 smoke grid artifact

[goshone]

Is anyone else getting these weird grid artifacts in your smoke sims? I just started diving into H12 and testing out some smoke simulations. After working on them a bit, I kept noticing very prominent grid artifacts in the viewport. I thought they would go away during the render, but they persist, and ruin the look quite a bit. It is minimized but still visible with higher resolutions. I went back to H11 and this doesn't seem to be a problem in that version. Is this something common among other users? I have attached a sample scene and image.

build_debris_fxTest_v001.hip

[old school]

Houdini12 has different defaults for Volume Filter and Volume Filter Width on Objects that will definitely cause the voxel cells to be visible for low resolution simulations. This was done to support sharper volume detail for high resolution volumes.

H12 uses Box Filter for the Volume Filter with a Volume Filter Width of 1 voxel. Very aggressive filter for coarse volumes.

In H11 it was Gaussian Filter and the Volume Filter width was 1.2 voxels if you want to duplicate for a much softer look.

Let us know if this helps.

[goshone]

Houdini12 has different defaults for Volume Filter and Volume Filter Width on Objects that will definitely cause the voxel cells to be visible for low resolution simulations. This was done to support sharper volume detail for high resolution volumes.

H12 uses Box Filter for the Volume Filter with a Volume Filter Width of 1 voxel. Very aggressive filter for coarse volumes.

In H11 it was Gaussian Filter and the Volume Filter width was 1.2 voxels if you want to duplicate for a much softer look.

Let us know if this helps.

Yes, this helps quite a bit in fact. That seems more along the lines of what I would expect from older versions, as well as from other software. It also seems that there needs to be a bit more care put into the sims to break up the regular-ness. After looking at the beautiful results from some of the pyro fx tab buttons, I realized that adding some noise to the source or the container (or both) produces very nice simulations even at low resolutions.

Another thing I noticed is that the volumes look much better WITHOUT any material assigned to it. When you attach the billowy smoke shader, it multiplies the density by 10 (by default) exaggerating the aliasing.

I will continue to investigate these and post any additional tips or info I glean from the process.

If there is any advice on solver or shader selection (smoke vs. pyro) and the pros/cons of each I would be very interested to hear.

[goshone]

So I have been trying further with this, on more of a real world shot. As much as I try, I am still unable to get rid of this grid issue. It is clearly visible in the viewport, and can minimize it as described above, but then the smoke becomes very blurry, and it still is perceptible. It has me wondering if I am simply doing something wrong. I initially thought it had to do with the source geo being only a poly shell and not a full volume. I closed the geo and used iso-offset to become a solid volume, but that did very little to fix the problem. The source is moving which can always be tricky because of substeps and stamping, but even emitted density becomes very striated with these grid lines.

So is my box simply too low resolution? The box in the image is about 125x75x125, which seems reasonable. I am just worried about the visible lines in the viewport, as they translate directly to the renders.

In this particular scene, which I am unable to share at the moment, I am using the smoke solver with some added noise applied to the source (density and velocity noise) and some turbulence in the container, both of which are aimed at moving the density around and blur these grid lines. Not having much luck though.

Any assistance would be appreciated.

[lukeiamyourfather]

It looks like the emission density is 1.0 or close to it which will lead to hard edges between voxels regardless of the software used. Breaking up the emission and motion will minimize the artifacts. Higher resolution will help as well but if you simply increase the resolution of the grid and change nothing else the problem will still be there to an extent.

Edited April 11, 2012 by lukeiamyourfather

[goshone]

True, it seems that it is a combination of techniques to get the most out of these solvers. The source of the volume is fairly important IMO, and the results are better when allowed to be moved around the container by velocity and forces rather than collisions or emissions. This can be seen with obvious grid artifacts around the emission point which get smoothed out over time during the simulation as the volume evolves within the container.

I may have had overestimated things a bit too much, and expected beautiful low res simulations straight out the gate. That being said, the performance from H12 is stellar, and that is without harnessing the GPU/openCL side of things.

There are a couple of render tests attached below, a lower resolution sim (v001c) which had some disappointing line artifacts, and another sim (v001d) with half the voxel size (so basically 4x the voxel count) which is drastically improved, although it may be hard to tell from the quicktimes. These have no material assigned, so that should only improve things, especially with some noise and adjusting the edge falloff.

Anyway, just some observations... still continuing to test out this awesome version release.

build_debris_fxTest_v001c_mantra.mov

build_debris_fxTest_v001d_mantra1.mov

[goshone]

Something that is throwing me off, whether it's a feature or a bug, is how the expanding volume container is displayed in openGL. I have included a playblast of my sim, which looks quite nice in the render so far. But looking at the chunky volume, I was very concerned of the final images. Seeing how the volume is nicely defined towards the beginning of the sequence, I think it is not actually adding addl voxels in GL preview, but expanding the existing slices at the beginning to describe the increasing volume. The best analogy I can give is this behavior is similar to metaballs display, where the polycount is initialized at some point, but as your metaball mesh expands, it doesn't actually add more polys to account for this, resulting in a very coarse mesh as the size and complexity of the metaball cluster increases.

Hopefully that all makes sense.

build_debris_fxSetup_v008_playblast.mov

build_debris_fxSetup_v008_mantra.mov

[johner]

By default there's a limit of 128^3 for the 3D textures used for the GL preview, so once you exceed that the data is quantized, leading to artifacts. You can increase that limit (or remove it altogether) under the Texture tab of the Display Options. Setting HDR Textures to 16b HDR Only can give smoother volume display as well.

[old school]

You also have per volume control over the viewport resolution with the Volume Visualization SOP.

Append a Volume Visualization SOP to your volumes, enable the Max Vis Res parameter and override the maximum resolution to whatever you want.

[eirik.m]

Hi Goshone

Have you tried to turn the filter on the Blackman with a value about 1.5 to 2.0?

It improved my render greatly

[goshone]

Hi Goshone

Have you tried to turn the filter on the Blackman with a value about 1.5 to 2.0?

It improved my render greatly

I just tried this, but things came out a bit toooo blurry for my tastes. It could help in some cases tho. I think it is good to know there are options to get the look you want. After further testing I have found that this artifact is most noticeable where heavy density meets no density, like the leading edge of an expanding billow of smoke. Higher resolutions seem to help, as do varying up the velocity (via fields or turbulence) as much as you can without breaking the look you are going for.