Reasons for Fireflies/Hotpixels

[indexofrefraction]

Hi…
I couldnt find a thread discussing the Reasons for Fireflies/Hotpixels,
so i try to get help here…

I work on a simple interior scene, a small room of 20m2 with windows on 2 sides.
The problem is, i get a lot fireflies/hotpixels!
The dont show first but after 1000/2000 samples they get more and more.
after 4k samples the normal noise is gone but i have a hotpixel noise instead.

This is annoying because it is a basic scene which would not be a problem for Vray,
and from what i see more samples would make it even worse.

So the question : WHAT ARE THE GENERAL REASONS FOR HOTPIXELS ?

I dont use special materials, just diffuse, glossy, specular
i have 2 large windows, which are covered by 2 large lights (outside the frame)
no environment
i have some lights inside the room too, so there is plenty of light !
just simple direct lightning with gi_diffuse, too

i tried:
using hotpixel removal in camera (value down to 0.25) -> has an effect, but does not eliminate the source of the problem
replacing mesh lights (with emitter material) by regular lights -> no big effect
deleting the window glass -> no big effect
(but i noticed that the glass swallows a lot of light!, after removing it ineeded to reduce the outside lights power a lot)
using environment instead of the window lights -> no big effect
raising the sample rate for all the interior lights (smaller light = bigger rate) -> no big effect
lowering bump values -> no big effect

now i m out of ideas … :/
anybody with tips what to do?
are glossy specular / roughness values a factor, maybe? or glasses with fake shadows on?

[kavorka]

I don't like direct light personally. Try changing it to PMC and turning the caustic blur to 1. PMC is a much better mode for reducing fireflies.

[dsyee]

I almost always set hotpixel removal to 0. In fact, I haven't run into a situation yet where setting it to > 0 improves things.

Fireflies are generally caused by having a large majority of paths ending in low light and a small minority of paths ending in bright light. For example if 99% of paths are occluded (end by hitting a non-emitting object), and 1% end at a small bright light, then you will probably get fireflies, because only 1 out of every 100 samples is actually coloring the pixel. Caustics can also produce this result because they refract light rays.

For caustics, you can a) use PMC, b) turn on fake shadows, and/or c) increase caustic blur to help fix it.

For fireflies in interiors, using a portal material on windows (see manual) can help, and also using larger, flat meshes for emitters. I don't know what you mean by "regular lights" because Octane (standalone) only supports mesh emitters (in addition to Daylight & Environment).

Maybe if you post an example, people can give you more specific advice?

[grimm]

You can also use the render region to help clean up problem areas.

[slepy8]

I think we misunderstood the question. Which is really important.

What is the real reason of fireflies??

Not: How to overcome the problem, how to go around.. THE REASON?!

This is the place where Otoy Octane Developers should give us some hints: what's the physical reason, how do light rays go around the scene leaving us with those shitty white spots??

There are some ways to go around, but undrstanding the problem is the key!!

So.. Otoy team??

[indexofrefraction]

about real lights… i m working on cinema4d, where you can put an emitting material on geometry…. or use a light object.
but thinking about it now it maybe is not a difference for octane in the core…. (i'll ask aoktar)

by the way i found a trick today that helped me…
i had 2 renders of the same camera…. (i did a lot of tests to reduce flies)
if you put these as layers into photoshop and select "darker color" as blend effect for the upper layer
all of the not exactly matching flies are eliminated !
needs 2 renders with slightly different settings, but i have a very clean output now.

but as also said above…
to really understand the reasons for flies would help to adapt the setup/materials/lights so that they dont even occur !
right now i had the problem that octane is very fast for the first impression of an image
but in the end i would have been faster getting a clean image with vray

btw, it would also be nice to know a bit more on how "hotpixel removal" works
does it blur the image? does it resample if a pixel is too bright next to the neighbours?

any inputs (specially of the octane team) are highly appreciated

[grimm]

Fireflies are, at least as far as I know, a side-effect of the path-tracing algorithm. All path tracing renderers have this problem, they just handle it differently. I remember a lot of conversations about it on the Indigo forums when Indigo was first started. Indigo and Octane are very much related as Radiance worked on Indigo before he started Octane. Fireflies are easier to handle on a CPU renderer than on a GPU one. So what causes the algorithm to do this? Small light sources, glossy surfaces with a lot of self reflection, or like dsyee said above.

[kavorka]

We use neat video, an after effects plugin that really really helps clean up images after the fact. I believe neat image for photoshop does the same. Check it out, it may work better than you re rendering a second pass.

It's not a replacement for letting an image cook, but full interiors can take a very long time to clean up and sometimes will still have a decent amount of noise after 20k samples. These plugins help clean it up without taking away quality

[kavorka]

oh, one more thing, those fireflies are often just 1 pixel in size. So if you increase the resolution of your image, they will be smaller and less noticeable. I often render at 2x my output I want because when I resize down, the image looks better than if I just rendered at that resolution from the beginning.

[dsyee]

Fireflies are generally caused by having a large majority of paths ending in low light and a small minority of paths ending in bright light. For example if 99% of paths are occluded (end by hitting a non-emitting object), and 1% end at a small bright light, then you will probably get fireflies, because only 1 out of every 100 samples is actually coloring the pixel. Caustics can also produce this result because they refract light rays.

What part of this does not make sense? Let me know and maybe someone can explain better. Fireflies are not a "mystery."

If you understand this, then you can take steps to eliminate them. But it may not always be possible.