OSL physical camera

[baltort]

I've got an OSL camera working with an arbitrary texture defining the shape of the aperture here: viewtopic.php?f=36&t=68775

As usual, I've developed it for Octane in Lightwave, but it should work in other hosts with little or no tweaking.

Let me know how you get on.

Cheers,

J.

[milanm]

I've got an OSL camera working with an arbitrary texture

No you don't. Show me a clean render done with this.

When you redirect(?) half of the samples you create holes in the render. And because the whole point of optimizing a pathtracer is to avoid hitting the same spot twice, those holes never get filled. The end result of that is that your render never gets clean + you loose 1 stop (I actually like that, because exposure doesn't go below 0.001 so an ND filter would be nice).

So, compared to this, the method of placing geometry in front of the camera mentioned earlier is actually much more efficient. Plus, you get "cat eye" bokeh as a bonus.

You could try to do it like we used to fake dof back in the day, by shifting the position of the camera and constraining the movement (to the bright pixel areas on the texture), while also keeping the focus point static (like a target). That way you don't loose any samples.

Oh, btw, out of curiosity, why did you build a fish-eye camera from scratch when we already had the one made by Roeland (Overview of OSL thread) fully working since November last year? Anything wrong with it?

Cheers
Milan

[baltort]

Hi Milan,

I'm not sure I follow your logic. Setting a ray direction to null is a quick way to create a zero valued ray sample (it's in the Octane OSL documentation). That is exactly what we want to represent any ray that is blocked by the aperture. I don't see any 'holes'.

Here's a quick tech test. I'm not sure what your threshold is for 'clean', but this is getting progressively less noisy.

Cheers,

James

[Jolbertoquini]

Hi Milan,

I'm not sure I follow your logic. Setting a ray direction to null is a quick way to create a zero valued ray sample (it's in the Octane OSL documentation). That is exactly what we want to represent any ray that is blocked by the aperture. I don't see any 'holes'.

Here's a quick tech test. I'm not sure what your threshold is for 'clean', but this is getting progressively less noisy.

star aperture test 2.jpg

Cheers,

James

Looking beautiful hehehe Nice work James!