ywu40 / gdconvnet

GDConvNet

This repo contains the official training and testing codes for our paper:

Video Frame Interpolation via Generalized Deformable Convolution

Zhihao Shi, Xiaohong Liu, Kangdi Shi, Linhui Dai, Jun Chen

Accepted in IEEE Transactions on Multimedia, 2021.

[Paper]

Prerequisites

• Python == 3.7.6
• Pytorch == 1.2.0
• Torchvision == 0.4.0
• Pillow == 7.1.2
• Numpy == 1.18.1

Quick Start

Dataset

viemo_septuplet is used as our training and testing dataset. Please download and unzip it somewhere on your device. Then change the training and testing directory in ./configs/config.py.

Testing

We have released the model weight in ./modeldict/, you can directly use it to do the evaluation. Using the command:

$ python3 eval.py

to start the testing process.

Training

You can also choose to retrain the model, just use the command:

$ python3 train_full_model.py

to start the training process.

Cite

If you use this code, please kindly cite

@article{shi2020video,
  title={Video Interpolation via Generalized Deformable Convolution},
  author={Shi, Zhihao and Liu, Xiaohong and Shi, Kangdi and Dai, Linhui and Chen, Jun},
  journal={arXiv preprint arXiv:2008.10680},
  year={2020}
}