-
New (Experimental) Features:
- Radiance Caching (RC):
- If RC is enabled, the guiding structure (i.e.,
Field
) learns an approximation of multiple radiance quantities (in linear RGB), such as outgoing and incoming radiance, irradiance, fluence, and in-scattered radiance. These quantities can be queried using theSurfaceSamplingDistribution
andVolumeSamplingDistribution
classes. RC support can be enabled using theOPENPGL_EF_RADIANCE_CACHES
CMake option. Note: Since the RC quantities are Monte-Carlo estimates, zero-value samples (ZeroValueSampleData
) that are generated during rendering/training have to be passed/stored in theSampleStorage
as well.
- If RC is enabled, the guiding structure (i.e.,
- Guided/Adjoint-driven Russian Roulette (GRR):
- The information stored in radiance caches can be used to optimize
stochastic path termination decisions (a.k.a. Russian roulette) to
avoid a significant increase in variance (i.e., noise) caused by
early terminations, which can occur when using standard
throughput-based RR strategies. We, therefore, added to example
implementation for guided
(
openpgl::cpp::util::GuidedRussianRoulette(...)
) and standard (openpgl::cpp::util::StandardThroughputBasedRussianRoulette(...)
) RR, which can be found in theopenpgl/cpp/RussianRoulette.h
header.
- The information stored in radiance caches can be used to optimize
stochastic path termination decisions (a.k.a. Russian roulette) to
avoid a significant increase in variance (i.e., noise) caused by
early terminations, which can occur when using standard
throughput-based RR strategies. We, therefore, added to example
implementation for guided
(
- Image-space guiding buffer (ISGB):
- The ISGB can be used to store and approximate per-pixel guiding
information (e.g., a pixel estimate used in guided Russian
roulette). The ISGB class
(
openpgl::cpp::util::ImageSpaceGuidingBuffer
) is defined in theopenpgl/cpp/ImageSpaceGuidingBuffer.h
header file. The support can be enabled using theOPENPGL_EF_IMAGE_SPACE_GUIDING_BUFFER
CMake option.
- The ISGB can be used to store and approximate per-pixel guiding
information (e.g., a pixel estimate used in guided Russian
roulette). The ISGB class
(
- Radiance Caching (RC):
-
API changes:
pgl_direction
: A new wrapper type for directional data. When using C++pgl_direction
can directly be assigned by and topgl_vec3f
.pgl_spectrum
: A new wrapper type for spetral (i.e., linear RGB) data. When using C++pgl_spectrum
can directly be assigned by and topgl_vec3f
.SampleData
:- New enum
EDirectLight
flag that identifies if the radiance stored in this sample comes directly from an emitter (e.g., emissive surface, volume, or light source). direction
: Changes the typepgl_vec3f
topgl_direction
.
- New enum
ZeroValueSampleData
: This new structure is a simplified and more compact representation of theSampleData
struct representing a zero-value sample. It contains the following members:position
: The position of the sample (typepgl_point3f
).direction
: The incoming direction of the sample (typepgl_direction
).volume
: If the sample is a volume sample (typebool
).
SampleStorage
: To add, query, and get the number ofZeroValueSampleData
, the following functions were added.AddZeroValueSample
andAddZeroValueSamples
: These functions add one or multipleZeroValueSampleData
.GetSizeZeroValueSurface
andGetSizeZeroValueVolume
: These functions return the number of collected/stored surface or volumeZe1roValueSampleData
.GetZeroValueSampleSurface
andGetZeroValueSampleVolume
: Return a givenZeroValueSampleData
from either the surface or volume storage.
-
API changes (
OPENPGL_EF_RADIANCE_CACHES=ON
): When the RC feature is enabled, additional functions and members are available for the following structures:SurfaceSamplingDistribution
:IncomingRadiance
: The incoming radiance estimate arriving at the current cache position from a specific direction.OutgoingRadiance
: The outgoing radiance at the current cache position to a specific direction.Irradiance
: The irradiance at the current cache position and for a given surface normal.
VolumeSamplingDistribution
:IncomingRadiance
: The incoming radiance estimate arriving at the current cache position from a specific direction.OutgoingRadiance
: The outgoing radiance at the current cache position to a specific direction.InscatteredRadiance
: The in-scattered radiance at the current cache position to a specific direction and for a given HG mean cosine.Fluence
: The volume fluence at the current cache position.
SampleData
:radianceIn
: The incoming radiance arriving at the sample position fromdirection
(typepgl_spectrum
).radianceInMISWeight
: The MIS weight of theradianceIn
if the source of it is a light source, if not it is1.0
(typefloat
).directionOut
: The outgoing direction of the sample (typepgl_direction
).radianceOut
: The outgoing radiance estimate of the sample (typepgl_direction
).
ZeroValueSampleData
: -directionOut
: The outgoing direction of the sample (typepgl_direction
). -
API changes (
OPENPGL_EF_IMAGE_SPACE_GUIDING_BUFFER=ON
): When the ISGB feature is enabled, additional functions and members are available for the following structures:ImageSpaceGuidingBuffer
: This is the main structure for storing image-space, per-pixel guiding information approximated from pixel samples. -AddSample
: Add a pixel sample of typeImageSpaceGuidingBuffer::Sample
to the buffer.Update
: Updates the image-space guiding information/approximations from the previously collected samples (e.g., denoises the pixel contribution estimates using OIDN). For efficiency reasons, it makes sense not to update the buffer after every rendering progression but in an exponential fashion (e.g., at progression2^0
,2^1
,…,2^N
).IsReady
: If the ISGB is ready (i.e., at least oneUpdate
step was performed).GetPixelContributionEstimate
: Returns the pixel contibution estimate for a given pixel, which can be used, for example, for guided RR.Reset
: Resets the ISGB.
ImageSpaceGuidingBuffer::Sample
: This structure is used to store information about a per-pixel sample that is passed to the ISGB.contribution
: The contribution estimate of the pixel value of a given sample (typepgl_vec3f
).albedo
: The albedo of the surface or the volume at the first scattering event (typepgl_vec3f
).normal
: The normal at the first surface scattering event or the ray dairection towards the camers if the first event is a volume event (typepgl_vec3f
).flags
: Bit encoded information about the sample (e.g., if the first scattering event is a volume eventSample::EVolumeEvent
).
-
Optimizations:
- Compression for spectral and directional: To reduce the size of the
SampleData
andZeroValueSampleData
data types it is possible to enable 32-Bit compression, which is mainly adviced when enabling the RC feature viaOPENPGL_EF_RADIANCE_CACHES=ON
.OPENPGL_DIRECTION_COMPRESSION
: Enables 32-Bit compression forpgl_direction
.OPENPGL_RADIANCE_COMPRESSION
: Enables 32-Bit compression forpgl_spectrum
.
- Compression for spectral and directional: To reduce the size of the
-
Bugfixes:
- Numerical accuracy problem during sampling when using parametric mixtures.
-
Platform support:
- Added support for Windows on ARM (by Anthony
Roberts
PR17). Note:
Requires using LLVM and
clang-cl.exe
as C and C++ compiler.
- Added support for Windows on ARM (by Anthony
Roberts
PR17). Note:
Requires using LLVM and
- Api changes:
Device
addednumThread
parameter (default = 0) to the constructor to set the number of threads used byOpen PGL
during training. The default value of0
uses all threads provided byTBB
. If the renderer usesTBB
as well and regulates the thread count this count is also used byOpen PGL
.SurfaceSamplingDistribution
andVolumeSamplingDistribution
:- Added
GetId
function to return the unique id of the spatial structure used to query the sampling distriubtion.
- Added
Field
andSampleStorage
, addedCompare
function to check if the data stored in different instances (e.g., generated by two separate runs) are similar (i.e., same spatial subdivisions and directional distributions).Field
:- The constructor of the
Field
class now takes aFieldConfig
instead of aPGLFieldArguments
object. (BREAKING API CHANGE) GetSurfaceStatistics
andGetVolumeStatistics
functions are added to query statistics about the surface and volume guiding field. The functions return aFieldStatistics
object. Note, querying the statistics of aField
introduces a small overhead.
- The constructor of the
FieldStatistics
:- This class store different statistics about a
Field
, such as, number and size of spatial nodes, statistics about the directional distributions, and the times spend for full and separate steps of the lastUpdate
step. The statistics can be queried as a full string (useful for logging) or as CSV strings (useful for analysis and plotting). ToString
: Returns a string printing all statistics.HeaderCSVString
: Returns the CSV header sting with the names of each statistic.ToCSVString
: Returns the CSV value sting of each statistic.
- This class store different statistics about a
FieldConfig
:- This class is added to replace the
PGLFieldArguments
struct when using the C++ API. -Init
: the function initializes the parameters of theFieldConfig
(i.e., similar topglFieldArgumentsSetDefaults
). Additional parameters (deterministic
andmaxSamplesPerLeaf
) are introduced to enable deterministic behavior and to control the spatial subdivision granularity. -SetSpatialStructureArgMaxDepth
: this function can be called afterInit
to the the maximum tree depth of the spatial structure.
- This class is added to replace the
pglFieldArgumentsSetDefaults
: Adding two additional parametersdeterministic
andmaxSamplesPerLeaf
. (BREAKING API CHANGE)
- Tools:
- Added a set of command line tools which are build when enabling the
OPENPGL_BUILD_TOOLS
Cmake flag.openpgl_bench
: Tool to time different components ofOpen PGL
such as the full training of aField
or the querying (initialization) ofSamplingDistributions
.openpgl_debug
: Tool tovalidate
andcompare
storedSampleStorage
andField
objects or retrain aField
from scratch using multiple stored sets (iterations) of stored samples.
- Added a set of command line tools which are build when enabling the
- Optimizations:
- Spatial structure (Kd-tree) build is now fully multithreaded. This
improves training performance on machines with higher core counts,
especially when using
deterministic
training. - Kd-tree switched to use cache-friendlier
TreeLets
instead of singleTreeNode
structures.
- Spatial structure (Kd-tree) build is now fully multithreaded. This
improves training performance on machines with higher core counts,
especially when using
- Bugfixes:
Field
fixed some non-deterministic behavior when spatial cache does not receive any training data during a training iteration due to a large number of training iterations.- Removed legacy/broken support for
OpenMP
threading since there is a dependency toTBB
anyway. - Fixed build problems on (non-Mac)
ARM
systems.
-
Api changes:
PathSegmentStorage
:-
Removed support for splatting training samples due to the fact that knn-lookups have proven to be better. Therefore, the function attributes
splatSamples
andsampler
have been removed from thePrepareSamples
function. -
Added
PropagateSamples
method prepare and push samples to theSampleStorage
The goal is to replacePrepareSamples
,GetSamples
andAddSamples
.
-
Sampler
:- Removed since it is not used/needed anymore.
SurfaceSamplingDistribution
andVolumeSamplingDistribution
:- The usage of parallax-compensation is now connected to the guiding
distribution type. Therefore the explicit
useParallaxCompensation
parameter is removed from theInit
functions of theSamplingDistributions
. - Added
IncomingRadiancePDF
function that returns an approximation of the incoming radiance distribution. This PDF does not need to be related to the actual sampling PDF but can be used for Resampled Importance Sampling (RIS).
- The usage of parallax-compensation is now connected to the guiding
distribution type. Therefore the explicit
Field
:- Adding
UpdateSurface
andUpdateVolume
function to update/train the surface and volume field separately.
- Adding
SampleStorage
:- Adding
ClearSurface
andClearVolume
function to clear the surface and volume samples separately. This allows to wait until a specific number of samples is collected for the surface or volume cache before updating/fitting theField
.
- Adding
-
Deactivating/removing
OMP
threading support since it would still have a dependency onTBB
-
Bugfixes:
- Fixing bug causing crash during
Field::Update
when in previous iterations no volume or surface samples were present.
- Fixing bug causing crash during
- Bugfixes:
- Fixing bug introduced in
0.4.0
when usingApplySingleLobeHenyeyGreensteinProduct()
for VMM-based representations
- Fixing bug introduced in
-
Performance:
- Optimized KNN lookup of guiding caches (x3 speed-up).
- Optimized Cosine product for VMM based representations.
-
Dependencies:
- Removed the Embree library dependency for KNN lookups in favour of the header-only library nanoflann.
-
Adding ARM Neon support (e.g., Apple M1).
-
Fixing memory alignment bug for higher SIMD widths.
-
PathSegmentStorage
:- Fixing bug when multiple refracted/reflected events hit a distant source (i.e., environment map) by clamping to a max distance.
- Adding
GetMaxDistance
andSetMaxDistance
methods. - Adding
GetNumSegments
andGetNumSamples
methods.
-
Field
:- Stopped tracing a total number of spp statistic since it is not
really useful.
- Removed the
GetTotalSPP
function. - Removed the
numPerPixelSamples
parameter from theUpdate
function.
- Removed the
- Stopped tracing a total number of spp statistic since it is not
really useful.
Field
:- Added
Reset()
function to reset a guiding field (e.g., when the lighting or the scene geometry changed)
- Added
PathSegmentStorage
:- Fixed bug when using
AddSample()
- Fixed bug when using
-
Added CMake Superbuild script to build Open PGL, including all its dependencies.
The dependencies (e.g., TBB and Embree) are downloaded, built, and installed automatically. -
Added support for different SIMD optimizations (SSE, AVX2, AVX-512). The optimization type can be chosen when initializing the
Device
. -
Added support for directional quadtrees for the directional representation.
-
PathSegmentStorage
:- Added debug function
CalculatePixelEstimate
to validate if the stored path segment information represents the sampling behavior of the render (i.e., the resulting RGB value should match the pixel value the renderer adds to the framebuffer)
- Added debug function
-
SurfaceSamplingDistribution
:- Added support for guiding based on the product of a normal-oriented
cosine lobe and the incident radiance distribution:
(ApplyCosineProduct)
This feature is only supported for VMM-based directional distributions. Support can be checked withSupportsApplyCosineProduct()
.
- Added support for guiding based on the product of a normal-oriented
cosine lobe and the incident radiance distribution:
-
VolumeSamplingDistribution
:- Added support for guiding based on the product of a single lobe HG
phase function and the incident radiance distribution:
ApplySingleLobeHenyeyGreensteinProduct()
This feature is only supported for VMM-based directional distributions. Support can be checked withSupportsApplySingleLobeHenyeyGreensteinProduct()
.
- Added support for guiding based on the product of a single lobe HG
phase function and the incident radiance distribution:
- Initial release of Open PGL Features:
- Incremental learning/updating of a 5D spatio-directional radiance
field from radiance samples (see
Field
). - Directional representation based on (parallax-aware) von Mises-Fisher mixtures.
PathSegmentStorage
is a utility class to help keep track of all path segment information and generate radiance samples when a path/random walk is finished/terminated.- Support for guided importance sampling of directions on surfaces
(see
SurfaceSamplingDistribution
) and inside volumes (seeVolumeSamplingDistribution
)
- Incremental learning/updating of a 5D spatio-directional radiance
field from radiance samples (see
- Added C-API and C++-API headers
- C-API:
#include <openpgl/openpgl.h>
- C++-API:
#include <openpgl/cpp/OpenPGL.h>
and the namespaceopenpgl::cpp::
- C-API: