Quadratic video interpolation

Implementation of “Quadratic video interpolation”, NeurIPS 2019.

Paper, Project

Packages

The following pakages are required to run the code:

• python==3.8
• pytorch==1.5.1
• cudatoolkit==10.1
• torchvision==0.6.1
• cupy==8.6.0
• tensorboardX
• opencv-python
• easydict

Video

Demo

• Download pretrained model and put it in “./qvi_release”
• Download weights of PWC-Net and put it in “./utils”

• Put your sequence in “datasets/example” with structure as

  seq1
  --00000.png 
  --00001.png
  --... 
  seq2
  --00000.png 
  --00001.png
  --...

• Then run the demo:

  python demo.py configs/test_config.py

  The output will be in “outputs/example”. Note that all settings are in config files under the folder “./configs”.

Train

• Download the QVI-960 dataset for training and put it in the folder “datasets”
• Download the validation data which is a subset of the Adobe-240 dataset, and put it in the folder “datasets”
• Then run the training code:

python train.py configs/train_config.py

Test

More datasets for evaluation:

• Adobe-240 (full version)
• GoPro
• UCF
• Davis

You can use “datas/Sequence.py” to conveniently load the test datasets.

Please consider citing this paper if you find the code and data useful in your research:

@inproceedings{qvi_nips19,
    title={Quadratic video interpolation},
    author={Xiangyu Xu and Li Siyao and Wenxiu Sun and Qian Yin and Ming-Hsuan Yang},
    booktitle = {NeurIPS},
    year={2019}
}