/GMFSS

GMFlow based video frame interpolation

Primary LanguagePython

GMFSS

GMFlow based anime video frame interpolation


2023-04-03: We now propose GMFSS_Fortuna as a factual basis for training in GMFSS. Please use it. This item will not be updated!

2022-11-23: GMFSS_union —— The next generation of GMFSS with better performance.

2022-09-18: GMFupSS —— A faster GMFSS


Requirement

  • Python 3.10.7 or other versions
  • opencv-python
  • pytorch
  • numpy
  • cupy
  • pandas

Usage

  • Run the following command and view the result in the 'output' folder
python test.py

Benchmarks on anime test set

  • Switch to project GMFSS_Fortuna
  • Download ATD-12K and unzip it, rename atd_test2k to dataset
  • Put the dataset to the root of eval.py
  • Run eval.py and see result at 'loss.csv'
PSNR: 29.22 SSIM: 0.932 LPIPS: 0.048

Also, have a look on VideoFlow’s First VFI application, VFSS!

Reference

Optical Flow: GMFlow

Video Interpolation: SoftSplat-Full SoftSplat AnimeInterp EISAI ECCV2022-RIFE

Acknowledgment

This project is sponsored by SVFI Development Team