This is a personal fork of this implementation of PWC-net for optical flow estimation. For more information check out the original repository or read the paper.
In this fork I've refactored the structure so that the network is separated and can be imported on it's own. I've also added a video_inference.py function so that video's can be fed into the optical flow model.
After installing pytorch and the necessary dependencies with e.g. conda, the rest can be installed with
pip install -r requirements.txtOnce this is done the pretrained network weights can be downloaded with
./download.sh