Bayesian Collaborative Denoiser for Monte-Carlo Rendering
BCD allows to denoise images rendered with Monte Carlo path tracing and
provided in the form of their samples statistics (average, distribution
and covariance of per-pixel color samples). BCD can run in CPU (e.g.,
renderfarm) or GPU (e.g., desktop) mode. It can be integrated as a library
to any Monte Carlo renderer, using the provided sample accumulator to
interface the Monte Carlo simulation with the BCD internals, and comes
with a graphics user interface for designing interactively the denoising
parameters, which can be saved in JSON format and later reused in batch.
BCD has been designed for easy integration and low invasiveness in the
host renderer, in a high spp context (production rendering). There are
at least three ways to integrate BCD in a rendering pipeline, by either:
* Dumping all samples in a raw file, using the raw2bcd tool to generate
the rendering statistics from this file and then running the BCD using
the CLI tool.
* Exporting the mandatory statistics from the rendering loop in EXR
format and running the BCD CLI tool to obtain a denoised image.
* Directly integrating the BCD library into the renderer, using the
sample accumulator to post samples to BCD during the path tracing and
denoising the accumulated values after rendering using the library.
Paths come from the repository package-file index for the observed builds. They describe archive/package associations, not every file that will exist on a running system after maintainer scripts, alternatives, generated state, diversions, or installation choices.
Jede Zeile ist Paketindex-Metadaten für eine Version, Architektur, Suite und ein Repository. Namen, URLs und Größen stammen aus der Quelle; ein Link ist ein veränderbarer Abrufort, kein Weitergabanspruch von OpenFactory.
