Eulerian Video Magnification
This repository contains a reproduction of Eulerian Video Magnification for Revealing Subtle Changes in the World. It is meant as an accessible implementation with low complexity that can be used as an aid for studying the algorithm.
How to run
Video files are not included in this repository due to licensing.
If you want to reproduce the paper with reproduce_paper.py, you will need to download the videos from the author's website.
-
Install FFmpeg to allow video file reading
Linux (Debian based):sudo apt install ffmpeg
MacOS:brew install ffmpeg
Windows: follow this guide -
Install the python dependencies by running
pip install -r requirements.txt -
Reproduce the experiments from the paper by running
python reproduce_paper.py
If skvideo cannot find FFmpeg, set the path to FFmpeg explicitly as shown in this Github issue. Magnifying a 10-second video can take up to a minute on an average laptop. If you cannot view the video output file, try opening it with VLC media player.
File structure
| Name | Purpose |
|---|---|
| filters.py | Ideal, Buttersworth and author's IIR signal filters |
| pyramids.py | Image dimension agnostic Laplacian and Guassian pyramid building functions |
| motion_magnify.py | The main function used to create a motion or color magnified video. |
| reproduce_paper.py | Runs experiments for the video files hosted on the paper's website |
Results
| Subject | Authors results | This repo results |
|---|---|---|
| Baby breathing |  |
 |
| Shadow on a house |  |
 |
| Guitar A string |  |
 |
| Human pusle (speed difference due to fps conversion) |  |
 |