Visualising RAFT OpticalFlow

Visualizing with a GIF optical flow.

The optical flow is computed with RAFT architecture (also works with no GPU, just slower)

Settings

To start, be sure to create a folder results in your current working directory. Then, create a folder with the name of your video, in which to save the results of the computed optical flow. In this folder there must be the video named as final.mp4.

It should look like this

current_directory
├── results/
│   └─── video_name/
│         ├─── rename.sh
│         └─── final.mp4
├── runner_optiflow.sh
└── optiflow.py

After that, be sure to know how many frames your video is long, then change that value in the file runner_optiflow.sh. Also, change the --input tag in the file runner_optiflow.sh with the name of the video you will work on.

Compute optical flow

Now we are ready. Just launch

./runner_optiflow.sh

Creating optical flow GIF

After having computed the optical flow for each couple of successive frames, be sure to change in rename.sh the correct number of frames to be renamed, launch

./rename.sh

Now execute

ffmpeg -f image2 -framerate 30 -i flow%d.png -loop -1 flow.gif

to finally generate the GIF visualazing the optical flow.

This is an example GIF

Acknowledgments

This code is based on the original paper of RAFT: https://arxiv.org/abs/2003.12039