Material Optimizer (GUI)

Student

Saip Can Hasbay, 01428723, University of Vienna, [email protected] or [email protected]

Requirements:

• Python > 3.8
• CUDA:
  • GPU: NVidia driver >= 495.89 (Taken from mitsuba3 requirements)

Dependencies:

• Please refer to requirements.txt for required python packages

Installation

1. Clone this repository

2. (optional) Create a virtual environment:

   1. Create:
      • python -m venv c:\path\to\myenv (e.g. python -m venv cloned_repository_root\myenv)
   2. Activate your virtual environment:
      • Windows: path_to_your_venv\Scripts\activate (e.g. myenv\Scripts\activate)
      • Linux: source path_to_your_venv\bin\activate (e.g. source myenv\bin\activate)

3. Install dependencies

   0. (optional): Make sure pip is updated
      • python -m pip install --upgrade pip
   • pip install -r requirements.txt

Run

• If a virtual environment is in use, make sure to activate it (see Installation 2.2)
• Run either through terminal:
  • python material_optimizer_gui.py
• Or IDE to your liking
  • e.g. vscode: Click to run icon on top right corner of material_optimizer_gui.py

Optional data

• To retrieve additional mitsuba-data (e.g. test scenes provided from the mitsuba team), then run the following:
  • git submodule update --init --recursive

Tips

• Change mitsuba version by modifiying the mitsuba row in the requirements.txt file and following the installation steps.

• On Visual Studio Code we suggests using the Python extension by Microsoft.

• You can check the availability of CUDA with the following command:

  • nvcc --version
  • Which should respond with something like this:

    nvcc: NVIDIA (R) Cuda compiler driver
    Copyright (c) 2005-2022 NVIDIA Corporation
    Built on Wed_Jun__8_16:49:14_PDT_2022
    Cuda compilation tools, release 11.7, V11.7.99
    Build cuda_11.7.r11.7/compiler.31442593_0
    

• If CUDA is not available in your system, or you receive an error similar to this:

    AttributeError: jit_init_thread_state(): the CUDA backend is inactive because it has not been initialized via jit_init(), or because the CUDA driver library ("libcuda.so") could not be found! Set the DRJIT_LIBCUDA_PATH environment variable to specify its path.
  

  • Linux: You might fix the problem with adding the following lines (if CUDA is indeed installed in your system) to your .bashrc file:

    # cuda path
    export PATH="/usr/local/cuda-12.0/bin:$PATH" # set according to cuda version
    export LD_LIBRARY_PATH=/usr/lib/wsl/lib:$LD_LIBRARY_PATH
    

  • Otherwise you might navigate to the ./src/material_optimizer_model.py file and replace the mitsuba variant as demonstrated below. Although at the moment we don't support LLVM (CPU) variant of mitsuba, it still might help to run the application.

      # set mitsuba variant: NVIDIA CUDA
      # mi.set_variant(CUDA_AD_RGB) # replace this line with
      mi.set_variant('llvm_ad_rgb') # with this
    

• Dr.Jit Cache location: C:\Users\user\AppData\Local\Temp\drjit\

• Automatic UML Model generation. See also.

  • pyreverse -o svg --colorized -p MaterialOptimizer . (make sure "." refers to this repository location)

• Beware that under linux you may need to change some path/s in provided Mitsuba XML files. For example, the file under scenes/material-preview/translucent-principled-bsdf/scene-init-bunny.xml, beginning for line 80:

    <shape type="ply">
        <string name="filename" value="scenes\material-preview\meshes\bunny.ply"/> # change this
        <string name="filename" value="scenes/material-preview/meshes/bunny.ply"/> # with this
    	<transform name="to_world">
    		<scale value="13"/>
    		<rotate z="1" angle="130"/>
    		<rotate y="1" angle="70"/>
    		<rotate x="1" angle="-50"/>
    		<rotate z="1" angle="-50"/>
    		<rotate y="1" angle="-2"/>
    		<rotate x="1" angle="12"/>
    		<translate x="0.3" y="0.0" z="-0.5"/>
    	</transform>
        <ref id="object_bsdf"/>
    </shape>
  

• Under videos directory you may find simple optimization examples.

Known Issues

• After closing material-optimizer sometimes Dr.Jit warns about leaked variables. For example:

  drjit-autodiff: variable leak detected (4 variables remain in use)!
   - variable a1154237 (1 references)
   - variable a1154245 (1 references)
   - variable a1154246 (1 references)
   - variable a1154244 (1 references)
  

  • Unfortunately the problem still persists, since programatically clearning cache for Dr.Jit is not possible. Further analysis necessary.
• On mitsuba version 3.1.1 during optimization mitsuba - thus material-optimizer - might crash with the following error message:

  Critical Dr.Jit compiler failure: jit_optix_compile(): optixModuleGetCompilationState() indicates that the compilation did not complete succesfully. The module's compilation state is: 0x2363
  

  • The problem seems to be known and will be fixed in the upcoming versions. See Also