Skip to content

Latest commit

 

History

History
29 lines (25 loc) · 1.36 KB

SETUP.md

File metadata and controls

29 lines (25 loc) · 1.36 KB
Raw

Setup

  1. Install anaconda or miniconda
  2. Install git, then clone respository: git clone https://github.com/harskish/tlgan/
  3. Create environment: conda create -n tlgan python=3.9
  4. Activate environment: conda activate tlgan
  5. Install dependencies
    • On NVIDIA Ampere GPUs (3000 series) or newer:
      conda env update -f env_cu11.yml --prune
    • On older GPUs:
      conda env update -f env_cu10.yml --prune
  6. Setup submodules: git submodule update --init --recursive

CUDA setup

The networks (based on StyleGAN2) contain custom CUDA kernels for improved performance.

  1. Install CUDA toolkit (match the version in env_cuXX.yml)
  2. On Windows: install and open 'x64 Native Tools Command Prompt for VS 2019'
    • Visual Studio 2019 Community Edition contains the required tools

Interactive viewers (optional)

The interactive viewers (visualize.py and grid_viz.py) benefit in performance from having access to a version of PyCUDA compiled with OpenGL support

Windows

Install the included dependencies:
pip install bin/cuXXX/*

Linux

  1. Install CUDA toolkit (match the version in env_cuXX.yml)
  2. Download pycuda sources from: https://pypi.org/project/pycuda/#files
  3. Extract files: tar -xzf pycuda-VERSION.tar.gz
  4. Configure: python configure.py --cuda-enable-gl --cuda-root=/path/to/cuda
  5. Compile and install: make install