riying / posnet

Pytorch Implementation of "PoSNet: 4x video frame interpolation using position-specific flow"

PoSNet

Pytorch Implementation of "PoSNet: 4x video frame interpolation using position-specific flow"

This method took 3rd place at AIM 2019 Video Temporal Super-Resolution Challenge

Related Work

In our work, we used the code of PWCNet and SpyNet. Thanks to Simon Niklaus for releasing the implementations.

Environment

python 3.7
pytorch 1.0.0

Prepare data

Test

Download Pre-trained models: [download] and place them in the './models' folder.

Example:

python main_middle.py --cuda --eval  # to interpolate middle frame
python main_side.py --cuda --eval  # to interpolate side frames

Training

To train your own model, you should prepare dataset using prepare_data.py.

Example:

python main_middle.py --cuda

Anknowledgement

Contact

If you have any question, feel free to contact me at [email protected]