A Dockerfile template for Mitsuba 3.
For CUDA backend, using nvidia-docker is required. Please follow the NVIDIA Container Toolkit Installation Guide.
docker build -t mitsuba3-docker .
Set the container name and data volume paths(like scenes, output directory, etc.) and run it.
Example:
docker run --name "hello" -v $(pwd)/scene:/mitsuba3/scene -v $(pwd)/output:/mitsuba3/output -e NVIDIA_DRIVER_CAPABILITIES=all --gpus all -it mitsuba3-docker bash
NVIDIA_DRIVER_CAPABILITIES=all
is required if you want to use GPUs. Without this option, OptiX initialization error may occurs.
Before building the docker image, edit mitsuba.conf
. See Configuring mitsuba.conf section in Mitsuba 3 official documentation.
The default setup in this repository is:
"enabled": [
"scalar_rgb",
"scalar_spectral",
"llvm_rgb",
"llvm_ad_rgb",
"llvm_spectral",
"cuda_rgb",
"cuda_ad_rgb",
"cuda_spectral"
],
You might have to enable the AD variants even though you don't need any differentiable rendering features. Actually I got an ADScope
related error in mitsuba python package when the compilation of AD backends was disabled.
Parts of this dockerfile are based on mitsuba2-docker by hiroaki-santo.