Hello everyone!
I have an issue with noice where some frames in an image sequence throw this error:
Error code 134: Output buffer denoised_outputs is accessed at -63, which is before the min (0) in dimension 0
It seems to be roughly every 20th frame with no traceable pattern. I could not find anything about this error on the web and it seems to me like a noice bug.
This is the (shortened) command line that we use:
noice --input "masterLayer.beauty.1278.exr" --input "masterLayer.N_noice.1278.exr" --input "masterLayer.Z_noice.1278.exr" --input "masterLayer.variance.1278.exr" --input "masterLayer.diffuse_albedo_noice.1278.exr" --patchradius 4 --searchradius 7 --variance 0.7 --extraframes 2 --output "masterLayer.beauty_denoised.1278.exr"
This is the output we get:
noice 5.3.0.2 [f7602f75] - the Arnold denoiser
Using 40 threads.
Loading images...
Loading file "masterLayer.beauty.1278.exr".
["..."]
Loading file "masterLayer.diffuse_albedo_noice.1279.exr".
Using feature AOV 'diffuse_albedo' with filter 'gaussian_filter'
Using feature AOV 'N' with filter 'gaussian_filter'
Using feature AOV 'Z' with filter 'gaussian_filter'
Working with 2 frames at 1x1
Will denoise AOV "RGBA", using associated variance
Output file will be "masterLayer.beauty_denoised.1278.exr"
Frame numbering from metadata is not consistent, using default indices.
Start denoising (patch radius 4, search radius 7, variance 0.7)
Denoising RGBA
Error: Output buffer denoised_outputs is accessed at -63, which is before the min (0) in dimension 0
Any information and/or hints on this are much appreciated!
Thanks,
Fabian
Hi Fabian/Jean Michel
Could you send some of the original frames to support, so we can investigate?
It could be an image with a very small data window
Hi Stephen!
Thanks for answering!
I have to ask production first before I can send you anything. Working on that.
I checked the frames and it is true that the sequence starts and ends in some black frames. The data window on these is 0x0 for the beauty, diffuse_albedo_noice and N_noice. Variance and Z_noice have a constant data window as big as the resolution.
There are some frames failing that have some data in the data window, but I think that can be because of the "extra frames" pulling in some frames with a data window of 0x0.
Example:
Working with 3 frames at 49x35
Will denoise AOV "RGBA", using associated variance
Output file will be "masterLayer.beauty_denoised.1286.exr"
Frame numbering from metadata is not consistent, using default indices.
Start denoising (patch radius 4, search radius 7, variance 0.7)
Denoising RGBA
Error: Output buffer denoised_outputs is accessed at -15, which is before the min (0) in dimension 0
Hope this helps for now
Fabian
Ok, then maybe I can reproduce the problem by setting up some EXRs with data windows like that...
Noice just crashed with the same error for me. The images I was running on sound similar to ones mentioned about.
To run:
./noice -patchradius 3 -searchradius 9 -variance 0.5 -ef 2 -f 20 -i 0030_lighting_masterLayer_specular_v008.0001.exr -i 0030_lighting_masterLayer_specular_v008_1.0001.exr -i 0030_lighting_masterLayer_specular_v008_default_1.0001.exr -o 0030_lighting_masterLayer_specular_v008_denoised.0001.exr
The output:
noice 5.4.0.0 [b7d9819c] - the Arnold denoiser
Using 40 threads.
Loading images...
Loading file "..."
...
Working with 3 frames at 1x1
Will denoise AOV "specular", using associated variance
Output file will be "0030_lighting_masterLayer_specular_v008_denoised.0001.exr"
Start denoising (patch radius 3, search radius 9, variance 0.5)
Denoising specular
Error: Output buffer denoised_outputs is accessed at -63, which is before the min (0) in dimension 0
I have attached test files.
Thanks!
What's the data window size? Seems anything smaller than 64x64 is a problem for noice