Air Module

[mfoxdogg]

Hey All. A while ago i was wondering why even unbiased renders even feel not quite there. And i came to the conclusion of what is missing in every rendering tech and thats Air.

Light scatters and dissipates as it passes through air, which is why we have god rays and glare and heightened lightbounces in dusty places. What luxrender does and other renderers do is render in a vacum or fake it with volumetrics (which is contained in a box not true atmosphere) hence why we have glare and bloom filters these alone make the renderer bias.

With the air module such effects would add a huge degree of believability and make renders much more nice and not look like photoshoped photos and Lux would be 100% unbias. this would also help in out scenes especially buildings and the amount of scatter and warp that air does in small but enough to make things seem not real. Its these little things you notice when they are NOT there.

This air module would be where path/ray is passing through space so even light from object to object would be slightly scattered.

Please consider this proposal, i would do it but have 0 knowledge if unbiased rendering and Lux's sources thanks for reading

[Lord Crc]

The current volume integration interface is very limited when it comes to the effects you're after.

However, from what I can understand, the new extended interface jeanphi's working will allow for such effects. It should, if I understood correctly, allow for the volume integrator to define scattering events inside the volume, which could be used to simulate water droplets suspended in air for instance for nice realistic glare around street lights a damp evening. If you let the volume go from essentially -infinity to +infinity you can get a realistic atmosphere.

However I suspect you'll want to rely on analytical results for "far away" effects ("blue mountains"). I'm not sure how the new volume interface can deal with that?

[tomb]

There are tons and tons of papers on this subject. It's fairly straightforward to implement a "true" atmospheric model but it does come at a cost. For the brute force way of doing it (single scatter mode), it requires integrating along rays for attenuation and backscatter towards the eye and computing inscatter from the sun along the way (see for example http://nis-lab.is.s.u-tokyo.ac.jp/~nis/ ... html#pas96 for the brute force algorithm - I have implemented it once upon a long time ago). However, the effect of attenuation/scattering is noticeable only at a quite great distance so a lot of processing power is "vasted" for nearby hits. For most scenes the analytical model used by lux (and most other renderers) is "good enough" and very fast.

But of course there are quite lot of other papers describing (increasingly clever) ways of precomputing various parts of the algorithm to speed things up, it just needs someone with the spare time and skills to implement This one looks quite interesting http://nis-lab.is.s.u-tokyo.ac.jp/~egak ... sia10.html

EDIT: linked to wrong paper

[jeanphi]

Hi,

The new interfaces should allow such effects. You can actually already use volumes for attenuation. Scattering is going to require some more work.

Jeanphi

[SATtva]

You can actually already use volumes for attenuation.

Does that mean absorption events can be calculated at the camera plane, not only at geometry intersection (i.e. you can get light absorption when the camera is located inside the volume)? This wasn't the case with former interfaces.

[tomb]

Hi,

The new interfaces should allow such effects. You can actually already use volumes for attenuation. Scattering is going to require some more work.

Jeanphi

Sweet - do the new interfaces support non-homogeneous densities too? Both molecular and aerosol densities in the atmosphere are 1/exponentially distributed (so-called scale height).

[jeanphi]

You can actually already use volumes for attenuation.

Does that mean absorption events can be calculated at the camera plane, not only at geometry intersection (i.e. you can get light absorption when the camera is located inside the volume)? This wasn't the case with former interfaces.

Yes you can, from quite some time already, or maybe some bits were missing I don't remember, but it should have worked. Can you retest? However it requires at least one intersection to work currently (ie it won't work if the light is directly visible from the camera).

Jeanphi

[jeanphi]

Sweet - do the new interfaces support non-homogeneous densities too? Both molecular and aerosol densities in the atmosphere are 1/exponentially distributed (so-called scale height).

Yes and no Actually volumes already support non-homogeneous properties (not the "clear" volume though). But the current interfaces might not evaluate several points along the path. The volume integrators need more work for that.

Jeanphi

[SATtva]

You can actually already use volumes for attenuation.

Does that mean absorption events can be calculated at the camera plane, not only at geometry intersection (i.e. you can get light absorption when the camera is located inside the volume)? This wasn't the case with former interfaces.

Yes you can, from quite some time already, or maybe some bits were missing I don't remember, but it should have worked. Can you retest? However it requires at least one intersection to work currently (ie it won't work if the light is directly visible from the camera).

Understood, I'm sure you'll overcome this limitation. I'll test as soon as new Linux weeklies are ready, working on this.

[jeanphi]

Understood, I'm sure you'll overcome this limitation. I'll test as soon as new Linux weeklies are ready, working on this.

Sure, It'll be easier to attach volumes to lights than camera, but in the end every case should be handled.

Jeanphi