OpenImageDenoise

[B.Y.O.B.]

I've set-up a second image pipeline but it doesn't change the image in any way, even though it becomes very slow and something is happening.

In their example, they check for errors after the filter is executed, not before:
https://github.com/OpenImageDenoise/oid ... pi-example

When we did the BCD integration, we started luxcoreui with

Code: Select all

-D screen.tool.type IMAGE_VIEW

So the image only updates every 4 seconds.
Also, you have to open "Window -> Film Outputs editor", look for your OIDN-imagepipeline and click "Show". This will trigger the imagepipeline for the first time.
Then you have to resize the new window.
You can re-run the imagepipeline with the "Refresh" button on the top left.

[alpistinho]

In their example, they check for errors after the filter is executed, not before:
https://github.com/OpenImageDenoise/oid ... pi-example

The example in the repo itself was executing before. In any case I've put it after the filter execution and it didn't throw anything.

When we did the BCD integration, we started luxcoreui with

Code: Select all

-D screen.tool.type IMAGE_VIEW

So the image only updates every 4 seconds.
Also, you have to open "Window -> Film Outputs editor", look for your OIDN-imagepipeline and click "Show". This will trigger the imagepipeline for the first time.
Then you have to resize the new window.
You can re-run the imagepipeline with the "Refresh" button on the top left.

I did that but there are no filters showing on the "Current Film Outputs settings". The log lines I've added indicate that the code is running, however.

Code: Select all

[LuxCore][179.756] [IntelOIDNPlugin] Copying Film to output buffer
[LuxCore][183.828] [IntelOIDNPlugin] Copying Film to input buffer
[LuxCore][183.882] [IntelOIDNPlugin] Executing filter

Can you check my configuration file to see if I'm doing something wrong with it?

Code: Select all

film.width = 512
film.height = 512
scene.file = scenes/cornell/cornell.scn
##
#path.photongi.sampler.type = RANDOM
#path.photongi.sampler.type = METROPOLIS
path.photongi.photon.maxcount = 400000
path.photongi.photon.maxdepth = 4
path.photongi.visibility.lookup.radius = 0.15
path.photongi.direct.enabled = 0
path.photongi.indirect.enabled = 1
path.photongi.indirect.maxsize = 400000
path.photongi.indirect.lookup.radius = 0.075
path.photongi.caustic.enabled = 0
#path.photongi.debug.type = showdirect
##
#path.pathdepth.total = 4
#film.filter.width = 2
#tilepath.sampling.aa.size = 21
#tile.size = 16
#tile.multipass.enable = 1
#tile.multipass.convergencetest.threshold256 = 16
#tile.multipass.convergencetest.threshold.reduction = 0.5
#screen.tiles.converged.show = 1
#screen.tiles.passcount.show = 1
#screen.tiles.notconverged.show = 1
#screen.tiles.passcount.show = 1
#screen.tiles.error.show = 1
#lightstrategy.type = DLS_CACHE
##
film.imagepipelines.0.0.type = GAMMA_CORRECTION
film.imagepipelines.0.0.value = 2.2
film.imagepipelines.1.0.type = GAMMA_CORRECTION
film.imagepipelines.1.0.value = 2.2
film.imagepipelines.1.1.type = INTEL_OIDN
#film.imagepipelines.1.0.type = BCD_DENOISER
#film.imagepipelines.1.1.type = GAMMA_CORRECTION
#film.imagepipelines.1.1.value = 2.2
##
#film.imagepipelines.0.0.type = PATTERNS
#film.imagepipelines.0.0.index = 0
##
#film.outputs.1.type = RGB_IMAGEPIPELINE
#film.outputs.1.index = 0
#film.outputs.1.filename = normal.png
#film.outputs.2.type = RGB_IMAGEPIPELINE
#film.outputs.2.index = 1
#film.outputs.2.filename = denoised.png
#film.outputs.2.type = MATERIAL_ID_COLOR
#film.outputs.2.filename = cornell_MATERIAL_ID_COLOR.png
batch.haltspp = 5

[B.Y.O.B.]

film.imagepipelines.1.0.type = GAMMA_CORRECTION
film.imagepipelines.1.0.value = 2.2
film.imagepipelines.1.1.type = INTEL_OIDN

I would put the gamma correction after the denoising step, but otherwise, I can't find anything wrong.
I'll try to compile your branch on my machine.

[alpistinho]

film.imagepipelines.1.0.type = GAMMA_CORRECTION
film.imagepipelines.1.0.value = 2.2
film.imagepipelines.1.1.type = INTEL_OIDN

I would put the gamma correction after the denoising step, but otherwise, I can't find anything wrong.
I'll try to compile your branch on my machine.

I just noticed that I was assuming that the Film would be tonemapped, the filter is even set up for LDR, but maybe that's not true.
Since there is no tonemapper defined, is the input film HDR and luxcoreui just tonemaps before drawing on the screen?

[Dade]

film.imagepipelines.1.0.type = GAMMA_CORRECTION
film.imagepipelines.1.0.value = 2.2
film.imagepipelines.1.1.type = INTEL_OIDN

I would put the gamma correction after the denoising step, but otherwise, I can't find anything wrong.
I'll try to compile your branch on my machine.

I just noticed that I was assuming that the Film would be tonemapped, the filter is even set up for LDR, but maybe that's not true.
Since there is no tonemapper defined, is the input film HDR and luxcoreui just tonemaps before drawing on the screen?

A normal image pipeline has a tone mapper plugin at some point. The denoiser should be the very first plugin in an image pipeline and it will receive the image in HDR format.

Something like:

Code: Select all

film.imagepipelines.1.0.type = INTEL_OIDN
film.imagepipelines.1.1.type = TONEMAP_LINEAR
film.imagepipelines.1.1.scale = 1.0
film.imagepipelines.1.2.type = GAMMA_CORRECTION
film.imagepipelines.1.2.value = 2.2

[alpistinho]

I would put the gamma correction after the denoising step, but otherwise, I can't find anything wrong.
I'll try to compile your branch on my machine.

I just noticed that I was assuming that the Film would be tonemapped, the filter is even set up for LDR, but maybe that's not true.
Since there is no tonemapper defined, is the input film HDR and luxcoreui just tonemaps before drawing on the screen?

A normal image pipeline has a tone mapper plugin at some point. The denoiser should be the very first plugin in an image pipeline and it will receive the image in HDR format.

Something like:

Code: Select all

film.imagepipelines.1.0.type = INTEL_OIDN
film.imagepipelines.1.1.type = TONEMAP_LINEAR
film.imagepipelines.1.1.scale = 1.0
film.imagepipelines.1.2.type = GAMMA_CORRECTION
film.imagepipelines.1.2.value = 2.2

I've set the filter to HDR and used your suggestion, but saw no change.
I've then changed some data structures to mimic exactly what they're doing on the documentation and while the image does not look denoised, it is different. It has some weird stripe pattern

[Dade]

It has some weird stripe pattern

The problem is in this kind of loop:

Code: Select all

for (u_int i = 0; i < pixelCount; i += 3) {
        color[i] = pixels[i].c[0];
        color[i + 1] = pixels[i].c[1];
        color[i + 2] = pixels[i].c[2];
}

It should be something like:

Code: Select all

for (u_int i = 0; i < pixelCount; ++i) {
        const i3 = i * 3;
        color[i3] = pixels[i].c[0];
        color[i3 + 1] = pixels[i].c[1];
        color[i3 + 2] = pixels[i].c[2];
}

[alpistinho]

It has some weird stripe pattern

The problem is in this kind of loop:

Code: Select all

for (u_int i = 0; i < pixelCount; i += 3) {
        color[i] = pixels[i].c[0];
        color[i + 1] = pixels[i].c[1];
        color[i + 2] = pixels[i].c[2];
}

It should be something like:

Code: Select all

for (u_int i = 0; i < pixelCount; ++i) {
        const i3 = i * 3;
        color[i3] = pixels[i].c[0];
        color[i3 + 1] = pixels[i].c[1];
        color[i3 + 2] = pixels[i].c[2];
}

Thanks Dade!

Sorry for the stupid mistake, it's working now =)

[alpistinho]

Is there support for imagepipeline plugins to receive multiple films as input?

Looking at the ImagePipelinePlugin class it seems there is not, should it be extended to receive multiple Films?

EDIT:
I guess I could also hardcode the IntelOIDN class to calculate the needed AOV's internally

[Dade]

Is there support for imagepipeline plugins to receive multiple films as input?

Looking at the ImagePipelinePlugin class it seems there is not, should it be extended to receive multiple Films?

What do you mean ? A film can have one or more image pipelines and each image pipeline can have any number of plugins. So if you use the Oidn plugin multiple times, there will be multiple/different instances of your class and each one will be called with a different film (so I assume the answer to your question is no).