#624 · Primary category: Computer Vision

tfoptflow

computer-vision cvpr2018 deep-learning flying-chairs kitti-dataset motion-estimation mpi-sintel optical-flow pwc-net tensorflow

Optical Flow Prediction with TensorFlow. Implements "PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume," by Deqing Sun et al. (CVPR 2018)

Project last updated:08/01/19

GitHub Stars

530

Forks

132

Contributors

1

License

MIT

Why we included this project

Optical flow estimation produces a dense motion map between consecutive video frames, and it shows up in everything from object tracking to action recognition. This repo is a TensorFlow implementation of PWC-Net, the CVPR 2018 architecture, and it covers the whole workflow rather than just a frozen model: training code, evaluation, and inference are all here. The author deliberately avoided dynamically loaded CUDA ops, so it runs on Linux and Windows, and it supports multi-GPU and mixed-precision training if you want to retrain on your own data. Pretrained weights are included, and the reported results match or beat the original paper on the MPI-Sintel 'final' benchmark. If you need a reference implementation of a classic optical flow network to build on, this is a solid place to start.

Articles for this project

No articles for this project yet.

To suggest a topic or contribute an article, contact us.

Related projects in this category