CoolColJ's test renders

[abstrax]

here

it would seem like the highlight at the bottom of the glass sphere has moved to the top in Luxrender
I'm surprised how different all 3 look as far as specular highlights go

Indigo was much faster under artifical lights than in the Sun compared to Thea and Lux. That whole lit area under the water was rendered in 1 big sweep pretty much, instead of little "dots" in the other 2. The light is a 10x10x5cm mesh light

Thea, Indigo, and Lux, in that order

The attachment CCJ_Water Pillars_Shootout2.JPG is no longer available

Found it. First I was seriously impressed with the Indigo result compared to Lux and Thea, but have a look at the differences. Lux and Thea have different noise but similar light distribution while Indigo is completely different. Just look at the highlight on the right column in the back or the brightness of the big spot under the glass sphere...

I then had a closer look at the light source and bingo: The light source a box that is open at the bottom, and the normals of the faces point outwards, i.e. to the sides and up. Lux area lights emit only in normal direction, most of the light got trapped between the roof and the light. Only the light that was actually sent to the pool was either indirect light reflected by the roof and the columns or the small amount of direct light coming from the thin side faces. No wonder that it took ages to converge.

Thea obviously behaves the same, but Indigo seems to emit light from both sides of a face. To reproduce the Indigo rendering, I inverted the normals of the light and duplicated the side faces and inverted them again, i.e. the side polygons emit now in both directions and the big top face emits down.

After baking the scene for 30min I get this result, it's not fully converged yet, but not too bad either:

Water_Pillars_Light_LUX_0002.png (146.98 KiB) Viewed 652 times

I hope this helps.

Cheers,
Marcus

[CoolColJ]

ahhh, I thought it was an enclosed box, I must have deleted a face by mistake

But do you see that extra highlight at the top of the sphere? None of the other renders have that

[CoolColJ]

I let this one render to 1500+ samples, with MNCR of 10 million, using variance, and there are some definite sampling errors (hotspots) and rendering errors - extra specular highlights that don't show up in Thea and Indigo
Even the shape of the middle top highlight on the sphere is different each time, and different to Indigo/Thea.

Water_Pillars_Light_VAR.JPG (18.6 KiB) Viewed 611 times

I rerendered the revised scene in Thea and Indigo, and they pretty match each other as far as Specular highlights go and their shape, but Lux is definitely the odd one out.

Indigo is still faster than the others when it comes to how it renders the underwater caustics under artifical lights. Instead of patch by patch like in Lux, or a collection of dots, in Thea, Indigo does big broad, smooth strokes.

[abstrax]

ahhh, I thought it was an enclosed box, I must have deleted a face by mistake

But do you see that extra highlight at the top of the sphere? None of the other renders have that

The extra top highlight is real. It's an internal total reflection of the light. Indigo has that too, if you zoom in (removed roof, pool and water for speed reasons):

This is what you get in Lux:

Similar results, but Indigo is more converged already as it calculates almost 2x samples/s than Lux. Also keep in mind that Cindigo exports spheres as parametric spheres, while LuxC4D only exports triangle meshes. I.e. the sphere shape in Indigo is more accurate.

Cheers,
Marcus

[CoolColJ]

Yeah, but I made the sphere as a mesh, so it should be a poly model. It certainly is in Thea, and it renders the the scene exactly like Indigo as far as the specular highlights go.

same scene, all rendered for 35 mins - the Lux specular highlight shapes are definitely off. Maybe the super sampling that Indigo and Thea do is the main factor here? I don't think it should matter that much

As far as Lux settings go - pretty much the defaults, but Max rejections at 1000 (like Indigo), and mutation range at 2% of scene size, 14 pixels in this case ( 50% of default) - makes it smoother.
I did one normal, and one with "use variance" on. The later seems to render better as far as how the samples get put down - smoother broader strokes, but they get laid down in separate areas sequentially. Instead of splatting down caustics all over the place, with "use variance" on, it's much more orderly. Like each area gets worked on until it hits a certain level before moving onto the next.
But there be some accuracy issues with "use variance".... so it's best not to use it right now.

order - Thea, Indigo, Lux - normal, use variance

[CoolColJ]

man the diffuse surfaces with Low discrepency + bidirection render so smooth and clean, so quickly it' scary
Why doesn't Metropolis do the same?

both rendered for 10 mins each

BLD

BiD_LowDis_10mins.JPG (17.82 KiB) Viewed 553 times

BMLT

BiD_MLT_1mins.JPG (29.87 KiB) Viewed 553 times

I don't know what you guys did and how, but that's really impressive, considering something like Indigo with almost 3x the samples per pixel can't match!

[jeanphi]

man the diffuse surfaces with Low discrepency + bidirection render so smooth and clean, so quickly it' scary
Why doesn't Metropolis do the same?

By nature MLT produces spots, this is a by product of the small mutations. However we do try to even them out as completely as possible. It should be possible to improve that a bit more, but would require some noticeable architectural changes, maybe for a future version.

Jeanphi

[patro]

hi CoolColJ,
nice to see your test here like on indigo forum....
your test are as usual very interesting and constructive!!!
the render of the fish is awesome..... why do you not create something with it for the splash screen contest?
do you can share the fish with the maps?.. obviously if it is not a personal work of you....
thank any way!!
I stay tuned for other test!!

PS: obj or 3ds

[CoolColJ]

thanks and check your PM soon

[CoolColJ]

SLG render at some 30,000+ passes
The artifacts don't converge away