View on GitHub


A Machine Learning Global Illumination renderer based on PBRTv3

Deep Radiance Caching - Usage Info

Basic rendering

To use Deep Radiance Caching on a PBRT scene, set the integrator field to iispt (naming discrepancies are due to project name changes that are not reflected in the code).

Integrator "iispt"

The integrator current does not have additional parameters.

Rendering can be launched by using the main launcher bin/pbrt

bin/pbrt <scene.pbrt>

By default the output will be scene.out.

Additional command line arguments

Other command line arguments can be used to specify the output filename (from which the output file format is deduced), for example

bin/pbrt scene.pbrt out.pfm

The target rendering quality for Deep Radiance Caching can be set on the command line. To set 32 indirect illumination tasks, and 16 direct illumination samples per pixel:

bin/pbrt scene.pbrt --iileIndirect=32 --iileDirect=16

Reference mode can be used to generate training data from a scene (in case you want to re-train the network with custom data). To enable the reference mode, set the reference tiling subdivisions to anything greater than 0. Set the number of path tracing samples per pixel for the ground truth examples as well. Reference resume option will automatically detect previously half-finished reference generation and attempt to continue from where it was left. This means you can safely Ctrl-C a reference generation session without losing too much work. Set Reference Resume to 0 to force re-generating everything. The output of the reference data is in the out directory from where the scenefile was.

bin/pbrt scene.pbrt --reference=20 --reference_samples=4096 --reference_resume=1