Awesome

SKFlow: Learning Optical Flow with Super Kernels

This repository is the official implementation of our paper:

SKFlow: Learning Optical Flow with Super Kernels

Shangkun Sun, Yuanqi Chen, Yu Zhu, Guodong Guo, Ge Li

NeurIPS 2022

Requirements

The code is tested on PyTorch 1.10.0, Python 3.9.7. To install requirements:

pip install -r requirements.txt

Data Preparation

SKFlow uses the following datasets for training and evaluation:

• FlyingChairs
• FlyingThings3D
• Sintel
• KITTI
• HD1K

Datasets are suggested to be organized as follows:

├── datasets
    ├── Sintel
        ├── test
        ├── training
    ├── KITTI
        ├── testing
        ├── training
        ├── devkit
    ├── FlyingChairs_release
        ├── data
    ├── FlyingThings3D
        ├── frames_cleanpass
        ├── frames_finalpass
        ├── optical_flow

Training

To train the model(s) in the paper, run this command:

sh scripts/train.sh

Evaluation

To evaluate our model (e.g. on Sintel), run:

sh scripts/infer.sh

Pre-trained Models

Pre-trained models could be downloaded here:

• Pretrained Model for Sintel
• Pretrained Model for KITTI
• Pretrained Model on C+T

Acknowledgement

Parts of code are adapted from the following repositories. We thank the authors for their great contribution to the community:

• RAFT
• GMA
• ptflops