Skip to content

Latest commit

 

History

History
85 lines (56 loc) · 2.47 KB

install.md

File metadata and controls

85 lines (56 loc) · 2.47 KB
Raw

Installation

System Dependencies

  • Linux (test on Ubuntu 16.04 and 18.04) or Windows (test on windows 10)
  • CUDA 10.1, 10.2, or 11.0 with Nvidia GPU.
  • gcc in Linux (supports C++14 and tests on 7.5+ ) or MSVC++ (Visual Studio 2019, supports C++14) in Windows.
  • ffmpeg (ffprobe) test on 4.3.1+.

Linux (test on Ubuntu 16.04 and 18.04)

  • Download the CUDA >= 10.1, and set the CUDA_HOME into the system environments. This is important. Since in the next setup stage, it needs to get the CUDA_HOME For example,

    export CUDA_HOME=/usr/local/cuda-10.1

  • Make sure that the gcc support C++14. Since in the next setup stage, it needs to compile some c++ codes from sources. Here, we use gcc 7.5.

  • It needs ffmpeg. If you do not install it. You can run the followings

    apt-get install ffmpeg

Windows 10

Python Dependencies

Setup Python

Using Python 3.6+. You can use Anaconda Python 3.6+, or the native Python 3.6+. Whatever Python you have used, be sure to create a virtual environment firstly.

  • For Anaconda Python

    conda create -n venv python=3.6.6
conda activate venv

  • For Native Python

    pip install virtualenv
virtualenv  venv
source venv/bin/activate

Requirements and Install iPERCore

Install iPERCore by running the follows

python setup.py develop

It might takes 5-10 min. When you see Finished processing dependencies for iPERCore==0.1.1 in the console screen, it means you have installed the iPERCore successfully.

The details of requirements. Again be sure you have used a virtual environment to avoid the conflicts with your own python site-packages.

Download checkpoints and samples

Run the followings scripts to download checkpoints and samples (recommend).

sh assets/download.sh

Or manually download the checkpoints and samples (option).

Unzip all of them, and move them to the ./assets directory.