Learning to Estimate Hidden Motions with Global Motion Aggregation (GMA)

This repository contains the source code for our paper:

Learning to Estimate Hidden Motions with Global Motion Aggregation
Shihao Jiang, Dylan Campbell, Yao Lu, Hongdong Li, Richard Hartley
ANU, Oxford

Environments

You will have to choose cudatoolkit version to match your compute environment. The code is tested on PyTorch 1.8.0 but other versions might also work.

conda create --name gma python==3.7
conda activate gma
conda install pytorch=1.8.0 torchvision=0.9.0 cudatoolkit=11.1 -c pytorch -c conda-forge
pip install matplotlib imageio einops scipy opencv-python

Demo

sh demo.sh

Train

sh train.sh

Evaluate

sh evaluate.sh

License

WTFPL. See LICENSE file.

Acknowledgement

The overall code framework is adapted from RAFT. We thank the authors for the contribution. We also thank Phil Wang for open-sourcing transformer implementations.