Skip to content
an implementation of Video Frame Interpolation via Adaptive Separable Convolution using PyTorch
Python Shell
Branch: master
Clone or download
Latest commit 6460c47 Nov 19, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
images no message Sep 10, 2017
middlebury no message Jun 12, 2019
sepconv no message Dec 23, 2018
videos no message Aug 13, 2019
README.md no message Nov 19, 2019
benchmark.py no message Aug 13, 2019
download.bash no message Aug 8, 2018
run.py no message Nov 19, 2019

README.md

sepconv-slomo

News: Check our new CVPR 2018 paper on a faster and higher-quality frame interpolation method.

This is a reference implementation of Video Frame Interpolation via Adaptive Separable Convolution [1] using PyTorch. Given two frames, it will make use of adaptive convolution [2] in a separable manner to interpolate the intermediate frame. Should you be making use of our work, please cite our paper [1].

Paper

For a reimplemntation of our work, see: https://github.com/martkartasev/sepconv
And for another adaptation, consider: https://github.com/HyeongminLEE/pytorch-sepconv

setup

To download the pre-trained models, run bash download.bash.

The separable convolution layer is implemented in CUDA using CuPy, which is why CuPy is a required dependency. It can be installed using pip install cupy or alternatively using one of the provided binary packages as outlined in the CuPy repository.

If you plan to process videos, then please also make sure to have pip install moviepy installed.

usage

To run it on your own pair of frames, use the following command. You can either select the l1 or the lf model, please see our paper for more details. In short, the l1 model should be used for quantitative evaluations and the lf model for qualitative comparisons.

python run.py --model lf --first ./images/first.png --second ./images/second.png --out ./out.png

To run in on a video, use the following command.

python run.py --model lf --video ./videos/car-turn.mp4 --out ./out.mp4

For a quick benchmark using examples from the Middlebury benchmark for optical flow, run python benchmark.py. You can use it to easily verify that the provided implementation runs as expected.

video

Video

license

The provided implementation is strictly for academic purposes only. Should you be interested in using our technology for any commercial use, please feel free to contact us.

references

[1]  @inproceedings{Niklaus_ICCV_2017,
         author = {Simon Niklaus and Long Mai and Feng Liu},
         title = {Video Frame Interpolation via Adaptive Separable Convolution},
         booktitle = {IEEE International Conference on Computer Vision},
         year = {2017}
     }
[2]  @inproceedings{Niklaus_CVPR_2017,
         author = {Simon Niklaus and Long Mai and Feng Liu},
         title = {Video Frame Interpolation via Adaptive Convolution},
         booktitle = {IEEE Conference on Computer Vision and Pattern Recognition},
         year = {2017}
     }

acknowledgment

This work was supported by NSF IIS-1321119. The video above uses materials under a Creative Common license or with the owner's permission, as detailed at the end.

You can’t perform that action at this time.