DIFRINT [TOG20 / SIGGRAPH Asia19]

This is the test code reference implementation of Deep Iterative Frame Interpolation for Full-frame Video Stabilization [1], using PyTorch. This work proposes a full-frame video stabilization method via frame interpolation techniques, making use of a self-supervised deep learning approach. Should you make use of our work, please cite our paper [1].

Our paper can be found in the ACM Digital Library and arXiv.

Setup

We used the following python and package versions:

python==3.5.6 torch==1.0.0 cupy==4.1.0 pillow==5.2.0 numpy==1.15.2 matplotlib==3.0.0 pypng==0.0.20 opencv-contrib-python==4.1.0.25 CUDA==9.0

You may require to setup the correlation package for computing the cost volume module in PWC-Net. If required, please follow the instructions in vt-vl-lab/pwc-net.

Usage

You can run python run_seq2.py --cuda --n_iter 3 --skip 2 to obtain example results on a sample given in the data folder, which will be saved in the output folder.

By default, our experiments were done with 3 iterations and skip parameter of 2. This can be customized by adjusting the --n_iter and --skip options.

We also provide code for making .avi videos from output frames, and a reference code for quality metrics.

Supplementary video

Please refer to the supplementary video provided below (click thumbnail):

References

[1] @article{Choi_TOG20,
	author = {Choi, Jinsoo and Kweon, In So},
	title = {Deep Iterative Frame Interpolation for Full-Frame Video Stabilization},
	year = {2020},
	issue_date = {February 2020},
	publisher = {Association for Computing Machinery},
	volume = {39},
	number = {1},
	issn = {0730-0301},
	url = {https://doi.org/10.1145/3363550},
	journal = {ACM Transactions on Graphics},
	articleno = {4},
	numpages = {9},
    }

License

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

Annbless / DIFRINT