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