Skip to content


PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume frames are extracted for every consecutive pair of frames in a video. PWC-Net is pre-trained on Sintel Flow dataset. The implementation follows sniklaus/pytorch-pwc@f61389005.

CUDA 11 and GPUs like RTX 3090 and newer

The current environment does not support CUDA 11 and, therefore, GPUs like RTX 3090 and newer. For details please check this issue #13 If you were able to fix it, please share your workaround. If you need an optical flow extractor, you are recommended to use RAFT.

Set up the Environment for PWC

Setup conda environment.

# it will create a new conda environment called 'pwc' on your machine
conda env create -f conda_env_pwc.yml

Minimal Working Example

Activate the environment

conda activate pwc

and extract optical flow from ./sample/v_GGSY1Qvo990.mp4 using one GPU and show the flow for each frame

python \
    feature_type=pwc \
    show_pred=true \

Note, if show_pred=true, the window with predictions will appear, use any key to show the next frame. To use show_pred=true, a screen must be attached to the machine or X11 forwarding is enabled.


Please see the examples for RAFT optical flow frame extraction. Make sure to replace --feature_type argument to pwc.


  1. The PWC-Net paper and official implementation.
  2. The PyTorch implementation used in this repo.


The wrapping code is under MIT, but PWC Net has GPL-3.0