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
In our work, we used the code of PWCNet and SpyNet. Thanks to Simon Niklaus for releasing the implementations.
python 3.7
pytorch 1.0.0
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
To train your own model, you should prepare dataset using prepare_data.py.
Example:
python main_middle.py --cuda
If you have any question, feel free to contact me at fkdlzmftld@gmail.com