Panoramic Camera Awesome

[treddie]

The Panoramic Camera was a great addition to Octane (can't remember when it came on board). It's amazing how much slower custom lights are relative to their equivalent in an HDR environment. That is especially true if you model things like parabolic reflectors with light blocks for your lights like what you would have in a real-world studio. In that case, all light rays are really INDIRECT light rays, because no light makes its way out into the scene directly, without first bouncing off the light block and reflector.

This is probably old news to many:

I just did a test where I removed everything from the studio...the seamless along with any other things meant to show up in glossy reflections, and "baked" just the three studio lights to an .exr image for my HDRI. So I had just three lights on a black background as a result. Then, in my scene (studio + subject(s)), I removed the three modeled light meshes, and replaced them with the HDRI for my environment and I have to say that my render times are probably roughly 100 times faster as a result. Where you really notice the improvement is in the dark areas where light is so scarce. Using HDR, the moderate to well lit surfaces cleared up in about one hour as well as the dark areas, with specular transparent surfs somewhat longer. When using modeled lights (meshes), I might have to spend 130 hours to render a complex scene and STILL have noise in the shadows and dark areas. But with the HDR equivalent...just one hour, MAAAYYYBE two. The only exception, as indicated above, is with specular surfs with dispersion. They take longer, but STILL, maybe 10 times faster?...I am running a test right now to see.

Since I use the exact same studio setup for all my test renders, I only spent 12 hours building the HDR, which I can now use forever. What is really cool about this, is that that 12 hours is to get a CLEAN, relatively noise free HDR. That is only really necessary when specular and high-gloss surfs reflect the image of those lights. If you do not have to worry about those reflections, you can drop the HDR render time down to practically nothing because the noise in the HDR will not show up on anything.

The only limitation with this idea is that the lighting HDRI cannot have any occluding objects burnt in as well...You have to leave that up to modeled meshes for that, unless it is something simple like a studio lights's light block sitting right in front of the bulb. So if you have recessed ceiling lights for example, you would have to figure out how to integrate the HDRI so that it shines through the recesses.

[treddie]

Here are my time tests:

The results seem counterintuitive. Although specular w/ dispersion was what I would expect, all the others are reversed...Surfaces recieving adequate light take longer than the dark, shadow areas on the same surfs, with cast shadows on diffuse surfs taking the longest. I can only assume this means that the diffuse surface has a much higher poly count per unit area.

For this example:
Dark areas in brushed metal: 65 times faster
Brushed metal surfs with adequate light: 19 times faster
Diffuse drop shadow: 10 times faster
Specular w/ dispersion surfs: 6 times faster

My initial guesstimate of 100 hours was based on the dark areas of the brushed metal, down in the seam along the bottom. Overall performance is quite a bit lower than that but still incredible (See attached image).

The conclusion I can draw from this is that if you can bake your modeled emitters into an environment HDRI without changing the way those emitters interact with the environment, this is the way to go for ANY scene regardless of poly counts and total amount of pixels in all the textures. Naturally, for really fast GPUs, this may not be such a big deal. Although for animation, everything counts.

Any thoughts or crits are welcome!

[slepy8]

I was going to make this test myself. You was faster
Thanks for sharing!

[treddie]

Heheh. One of us is psychic. Not sure which one.

[treddie]

Update on the specular w/ dispersion times:

At 47 hours there is still noticeable noise in the specular/dispersion areas, although it is nicely countered in NeatImage. Since Octane's noise reduction rate decreases logarithmically, it is very difficult to say when specular w/ dispersion will clear up completely.

But NeatImage does a good job at the 47-hour noise level. So compositing the NeatImage result of the glass into the render looks very good.

Testing specular W/O dispersion now.

[stratified]

Since Octane's noise reduction rate decreases logarithmically, it is very difficult to say when specular w/ dispersion will clear up completely.

Logarithmically is a bit slow, you wouldn't like it The error for Monte Carlo sampling is square root which means to halve the error (noise) you would need to take 4 times as many samples.

cheers,
Thomas

[treddie]

Excuse me, you are definitely correct. I meant an EXPONENTIAL curve, y = x^.5. Because it drops off exponentially, instead of accelerating upward, the curve shape resembles a logarithmic curve except that it doesn't pass through zero. But the shape erroneously triggered the "logarithmic" signal in my brain instead of "exponential".

[treddie]

UPDATE:

In my test setup, Specular w/o Dispersion (16hours) matched the noise level for Specular w/ Dispersion (47hours). Still grainy, but something like NeatImage can work with that for the specular areas, with compositing in PShop.

[treddie]

I just deleted my two last posts relating to differences in normal map heights looking different between the two lighting methods. This was incorrect...I had inadvertently used a hires version for my normal map in my latest HDRI test. I should have used the lores version.

So all is still very consistent between mesh emitter and HDRI methods.

One very slight issue is getting the HDRI position of the lamps to match the positions of the mesh emitters exactly. I have a very slight difference between the two methods that does influence the scene in the reflections.

My method for getting the position of the HDRI correct was:
--1. Panoramic Camera Mode: Spherical
--2. Horizontal Field of View: 360deg
--3. Vertical Field of View: 180deg
--4. Center camera origin at scene origin (0, 0, 0), and target somewhere along -Z (ie., 0, 0, -3).
--5. When using the HDR background:
--------a. Set Environment Gamma = 1.0
--------b. Texture Power = 1.0
--------c. Scene Power = 100 as starting value.
--------d. Projection: Mesh UV

[treddie]

I am experimenting with a different method for placing the panoramic camera. I hope to have a graphic explaining it here later today.