Skip to content
This repository has been archived by the owner on May 11, 2023. It is now read-only.
AlphaAtlas edited this page Nov 25, 2019 · 8 revisions

Why would anyone want this, as opposed to installing VapourSynth and everything else manually?

  • The dependencies for the VSGAN and super_resolution scripts are impossible to package in PyPI or VSRepo, and difficult to install on Windows by hand. For example, cuDNN can't be redistributed, the super-resolution-zoo repo is HUGE, certain packages only work with certain CUDA versions, etc. It's all time consuming and frustrating to manually setup as a beginner, and many of the errors you'd run into provide almost no info. And, for my own use, it streamlines the process of setting all this up on new machines.

Why use VapourSynth instead of Python utilities like ESRGAN Scripts, Gigapixel, Video2X, etc?

  • VapourSynth has a huge library of sophisticated image/video processing filters, with development history going all the way back to 2000/2001 (if you count all the AviSynth filters and their ports). You can use these for temporal denoising, deblocking, etc. before feeding media to Super Resolution algorithms, or you can postprocess those images before writing them to disk.
  • VapourSynth is pipelined. So instead of having to, for example, convert videos to a sequence of images, process them, and then convert them back, all that conversion happens at once, with no quality loss, and nothing but the results written to disk.
  • VSEdit provides a nice GUI for previewing video and image processing.
  • It's all free!

Why make everything portable instead of installed?

  • I have a bad experience with CUDA and Python environments. Sometimes pip will fail to properly uninstall something, sometimes I'll uninstall the wrong version, etc. If something goes wrong, and everything's portable, it's often easier to just delete everything and setup a new Python environment, which is one reason why Anaconda has such a sophisticated environment switching system. That's why I started this personal project.

Linux?

  • All this setup is easier on Linux, as you can install/compile pretty much everything you need through pip, Anaconda and/or some command in a terminal. So an "installer" like this isn't really necessary, but maybe someone can make an install script...