This repository contains the PyTorch implementation of optical flow warping. The implementation is based on the TensorFlow implementation.
When reproducing the result of PWC-Net, I tried to warp the optical flow back to the first frame, while the warping function in the model provided by the author did not work (maybe it's because I made some stupid mistake on normalization).
Demo data is also leveraged from the original TensorFlow implementation.
python run.py
The recon.png is generated in the data
folder.