Fighting vs fireflies

Post your tests, experiments and unfinished renderings here.
marcatore
Posts: 94
Joined: Wed Jan 10, 2018 8:04 am

Fighting vs fireflies

Post by marcatore » Fri Apr 06, 2018 7:24 am

Hi Luxusers,
I'd like to share with you a scene I'm using to test Luxcore
https://we.tl/qCkfsoVwat

I'm finding difficulties to avoid fireflies.
Probably the room is not an easy room but I'd like to share with you if I'm making some big mistakes.

So, if you have some time to take a look to it, and you'll have some good suggestions I'll really appreciate.

In the link above you'll find the scene (should have all the textures I used), the IES file (I'm not sure it's packed inside the file) and the rendered image I had with the render options I used.
Consider that I've rendered it in 2hours with a QuadroK5200 and Xeon E5-1650@3.50Ghz

This is what I've achieved (but I'm not happy speaking about noise and fireflies)
test_luxcore1.jpg

User avatar
Sharlybg
Supporting Users
Posts: 642
Joined: Mon Dec 04, 2017 10:11 pm
Location: Ivory Coast

Re: Fighting vs fireflies

Post by Sharlybg » Fri Apr 06, 2018 7:58 am

Easy :

Luxcore now is a fireflies killer when you know how to use it.

1/ never go above 0.7 or 0.8 in material color

2/ use clamp method and metropolis for interior

3/ use the correct light strategegy for your project (power/log power/Uniform)

4/ Switch to bidirection in interior

5/ you have to light your scene like a pro photographer ( or you get noise like noob photographer).
Support LuxCoreRender project with salts and bounties

marcatore
Posts: 94
Joined: Wed Jan 10, 2018 8:04 am

Re: Fighting vs fireflies

Post by marcatore » Fri Apr 06, 2018 8:38 am

First of all thank you for your points.
Sharlybg wrote:
Fri Apr 06, 2018 7:58 am
1/ never go above 0.7 or 0.8 in material color
I did. At least the main white material has the default 0.7 white value
I'll double check for the others.
Sharlybg wrote:
Fri Apr 06, 2018 7:58 am
2/ use clamp method and metropolis for interior
I'm using clamp already. I've quite lowered the value trying to set it near the artificial lighting instead the natural one.
Sharlybg wrote:
Fri Apr 06, 2018 7:58 am
3/ use the correct light strategegy for your project (power/log power/Uniform)
As I have sun&sky and artificial lighting, I have choosen "log power" as suggested in the tooltip (and, following the explanation it seems correct, otherwise what should I choose?)
Sharlybg wrote:
Fri Apr 06, 2018 7:58 am
4/ Switch to bidirection in interior
I'll try. Have you some settings for bidir to suggest?
Sharlybg wrote:
Fri Apr 06, 2018 7:58 am
5/ you have to light your scene like a pro photographer ( or you get noise like noob photographer).
The room is an office near to mine where I'm working. If the room is empty, I can take a picture as reference. I'm quite sure that with a prosumer camera and without fill lights I can enlight correctly that room. Anyway I'll verify to have the lamps correctly set.

User avatar
B.Y.O.B.
Developer
Posts: 1329
Joined: Mon Dec 04, 2017 10:08 pm
Location: Germany
Contact:

Re: Fighting vs fireflies

Post by B.Y.O.B. » Fri Apr 06, 2018 9:12 am

Things that make this scene difficult to render:
  • Lots of point lights (around 80)
  • The point lights have their gain boosted by 10 in the light group. Note that the light strategy does not take the gain setting of light groups into account. All light groups should have gain 1 and white color at the start of the render, they are intended for getting different lighting situations out of a single render, not for adjusting lots of lights at once during the setup of the scene. These feature requests are related.
As suggested by Sharlybg you should try to render with Bidir, it should handle the massive amount of point lights better than Path.
And maybe you can consider turning the ceiling lamps off and only leaving a few on, this may lead to more interesting lighting in the scene (less uniform) and also makes the render faster.
marcatore wrote:
Fri Apr 06, 2018 8:38 am
I'll try. Have you some settings for bidir to suggest?
The defaults.
Right click -> Reset to Default Value.
The metropolis sampler settings in your file are all over the place, use the defaults here, too.
Support LuxCoreRender project with salts and bounties

User avatar
Dade
Developer
Posts: 1187
Joined: Mon Dec 04, 2017 8:36 pm

Re: Fighting vs fireflies

Post by Dade » Fri Apr 06, 2018 9:19 am

The 80 point light sources, with 80 mirrors near, are quite a killer. You should replace each "point light+mirror" with just a point light with an IES profile. It shouldn't be hard to find an IES profile for the type of lamps you are using, they are quite common. It will make an huge difference in term of fireflies.
Support LuxCoreRender project with salts and bounties

marcatore
Posts: 94
Joined: Wed Jan 10, 2018 8:04 am

Re: Fighting vs fireflies

Post by marcatore » Fri Apr 06, 2018 9:55 am

Good points...

Some quick feedback and point of discussion

- I've tried with bidir in a region and it seems that it performs better in this scenario. I'll rerender full image with it.

- about the point light and IES. They're already using a profile. You'll find in the linked zip file. I hope that the blend file keep the file linked. If not and you want to try it, relink with that I've provided.

- about mirror near the light. I'm sorry but a lamp has that mirror. The problem is that it's a reflective material or that it's a mirror with no glossy? Because the real lamp as reflective materials in the projector so I think that it's not a debatable thing. But if applying a glossy reflective material could help I will do (and it's more real actually).

- about the gain I'll bring it back to the default.

- about turn off some lights, sorry but I prefer to simulate a real scenario as much as I can , with less subjective modifications. Consider that the real room as more lights that mine scene. I agree that turning off someone will produce a more interesting lighting but I'd like to test with this kind of lighting, not so much different from the real one.

User avatar
Dade
Developer
Posts: 1187
Joined: Mon Dec 04, 2017 8:36 pm

Re: Fighting vs fireflies

Post by Dade » Fri Apr 06, 2018 2:37 pm

marcatore wrote:
Fri Apr 06, 2018 9:55 am
- about the point light and IES. They're already using a profile. You'll find in the linked zip file. I hope that the blend file keep the file linked. If not and you want to try it, relink with that I've provided.

- about mirror near the light. I'm sorry but a lamp has that mirror. The problem is that it's a reflective material or that it's a mirror with no glossy? Because the real lamp as reflective materials in the projector so I think that it's not a debatable thing. But if applying a glossy reflective material could help I will do (and it's more real actually).
Nope, you can have an IES file that define the light emission including the mirror. At that point, you have to remove the mirror because its interaction with the light source is already included in the IES profile. This is exactly why IES profiles exist, it is like a pre-processing step and the difference in term of fireflies and rendering time is like night and day. But you have to careful check what the vendor is including in the IES profile and what is not; especially for light simulation more than just artistic renderings.
Support LuxCoreRender project with salts and bounties

marcatore
Posts: 94
Joined: Wed Jan 10, 2018 8:04 am

Re: Fighting vs fireflies

Post by marcatore » Fri Apr 06, 2018 2:52 pm

This is a bidir render.
test_luxcore_1700samples_BiDir_4h20min.jpg
As suggested I've reset to the defaults the bidir options.

The point lights has IES file, I've deleted all the light groups so the lamps had gain set to 1 and all the other defaults.
The Light strategy is "Log Power". Is it correct for that scenario?

A general question to better understand how to set the final samples halt conditions.
I did a render region with 1700 samples and I had this kind of noise (good and fireflies free)
test_luxcore_1700samples_42min.jpg
Why when I did the full render image the 1700 samples give me that noisy result?
It seems that, with region render, the samples are more concentrated because the rendered area is smaller than the final image.
If it's correct, to get the same noise quality I should estimate how much % of area I had with region and then improve the number of samples to cover the full image. In numbers... for example... if 1700 samples is ok for a region of 25% I should set 1700*4 = 6800 samples for the full render.
Am I right?

marcatore
Posts: 94
Joined: Wed Jan 10, 2018 8:04 am

Re: Fighting vs fireflies

Post by marcatore » Fri Apr 06, 2018 3:01 pm

Dade wrote:
Fri Apr 06, 2018 2:37 pm
Nope, you can have an IES file that define the light emission including the mirror. At that point, you have to remove the mirror because its interaction with the light source is already included in the IES profile. This is exactly why IES profiles exist, it is like a pre-processing step and the difference in term of fireflies and rendering time is like night and day. But you have to careful check what the vendor is including in the IES profile and what is not; especially for light simulation more than just artistic renderings.
Understood and you're technically right but now...if you should render a light lamp that in real life has mirror in its projector how did you set the scene?
I mean that I want to see the light reflected inside the lamps.
I've in mind these possibilities:
- I delete all the point lights and the projector has an emission material in all the internal lamp faces.
- I delete all the point lights and i assign an emission material for the faces that represent the LED part of the lamp and I assign a mirror for the other internal faces.

have you another suggestions?

User avatar
Dade
Developer
Posts: 1187
Joined: Mon Dec 04, 2017 8:36 pm

Re: Fighting vs fireflies

Post by Dade » Fri Apr 06, 2018 3:21 pm

marcatore wrote:
Fri Apr 06, 2018 2:52 pm
Why when I did the full render image the 1700 samples give me that noisy result?
It seems that, with region render, the samples are more concentrated because the rendered area is smaller than the final image.
If it's correct, to get the same noise quality I should estimate how much % of area I had with region and then improve the number of samples to cover the full image. In numbers... for example... if 1700 samples is ok for a region of 25% I should set 1700*4 = 6800 samples for the full render.
Am I right?
Yes, this may be considered somewhat a bug or a counter intuitive behavior, the halt condition is always evaluated for the complete image even when rendering a small region. So if you render 50% of the image with an halt condition of 1700 samples is like rendering 100% of the image with an halt condition of 3400 samples.

This because to achieve an average of 1700 samples per all pixels when rendering only half of them, you have to render 3400 samples.
Support LuxCoreRender project with salts and bounties

Post Reply