Open denisri opened 8 months ago
With an up-to-date Conda environment, I have been able to install pytorch
and torchvision
without conflict. I check them but I did not test their use.
We can install pip packages from conda packages using scripts that are called after installation. This is not the recommended way of doing things but it works for installation. It will make uninstallation difficult because we can only remove the pip package without knowing if another package needs it or if the user decided to manually install the package.
For dracopy
, there is a draco
package in conda-forge. However, the package contains symlinks for the commands draco-decoder
and draco_encoder
but they are not installed. Therefore, for my installation only draco_decoder-1.5.7
and draco_encoder-1.5.7
are available in PATH. Despite this bug, there is more in the Conda package than in the pip one (e.g. CMake files). I will make a dependency on the conda package.
For webp
there is a conda package pywebp
. I will make a dependency on the Conda package.
For pygltflib, I will use the pip version.
Pip dependencies are now supported and checked software had been included in casa-distro source and in test packages.
I have been able to create a conda package for virtualgl. I added code to create a repository called brainvisa-forge
on setup and to create a brainvisa-virtualgl
package in that repository. This repository is added in the list of conda repositories. Therefore brainvisa-virtualgl
can be used as a dependency. I added a brainvisa-forge
specific list of dependencies that we can define for our packages.
There is a package on conda-forge containing xvfb : xorg-x11-server-xvfb-cos7-x86_64
It is specific to an architecture (x86_64
) and a low level libraries (cos7
for CentOS 7). These packages are not installed in the standard path. Therefore I modified bv_env
in brainvisa-cmake to add this specific directory in the PATH
in conda environment.
The Conda distribution is starting to work. However some optional packages are missing for a full-featured brainvisa distribution. This ticket will list missing optional packages and try to find solutions.
torch
andtorchvision
modules, which exist on condaforge but apparently cause installation conflictspygltflib
module for .glb binary variant of GLTF format. This module doesn't exist as a condaforge package, but may be installed using pip. Is it OK to mix conda and pip ?dracopy
module for GLTF/GLB meshes compressionwebp
module for GLTF/GLB textures compressiongltf-transform
tool for GLTF/GLB meshes compressionvirtualgl
for off-screen 3D rendering and other usesxvfb
for off-screen rendering (headless Anatomist...)