A python implementation MIT's Eulerian Video Magnification algorithms from both the Siggraph 2012 paper (Hao-Yu Wu, Michael Rubinstein, Eugene Shih, John Guttag, Frédo Durand and William T. Freeman) and also the "new-and-improved" phase-based approach from the 2013 paper (Neal Wadhwa, Michael Rubinstein, Frédo Durand and William T. Freeman). The code presented here is a python implementation of the matlab sources provided by MIT, which are used under license with all credit going to the authors of these algorithms; additionally, the linear implementation leans heavily on https://github.com/LabForComputationalVision/pyrtools for generating Laplacian pyramids.
The original papers can be found here (along with more recent developments) http://people.csail.mit.edu/mrub/vidmag/ and they make for an excellent read.
Installation: Installing into a virtualenv is recommended. At the command prompt, enter:
pip install -e .
To reproduce the results for the Siggraph 2013 paper:
- Download the source videos (1.3GB) from the
project web page into a directory called
sample. - Type
reproduce_results_siggraph13to reproduce the results in the paper.
There is also a little tkinter UI for demo purposes, which can be launched from the shell like this:
phase_amplify_app
The gifs below are the MIT file 'car_engine.avi' before & after processing with the phase-based algorithm:
Note that OpenCV 4.x is required (should be compatible with 3.x versions too).