Second project given as part of the following course: Unsupervised learning, fall 2017-2018, MSc MVA, ENS Paris-Saclay.

Team: Vincent MATTHYS, Pirashanth RATNAMOGAN, Othmane SAYEM

The instructions are given in project2-cachan17.pdf

Consider the problem of clustering face images of multiple individuals acquired with a fixed pose and varying illumination. The goal is to cluster the faces, each of one belonging to a given individual, as shown in the following figure, extracted from Vidal et al, 2016

Consider a video sequence in which a camera observes n distinct points. The goal is to cluster the points, each of one have a relative motion, as shown in the following figure, extracted from Vidal et al, 2016:

The data are available online, thanks to the The Vision, Dynamics and Learning Lab of John Hopkins University.

If you already have downloaded data, please skip the download subsection and proceed to the creation of symbolic link as suggested in symbolic link subsection

To download the data as required by the notebooks, you have to proceed as follow:

wget -O data/ExtendedYaleB.mat "http://www.vision.jhu.edu/gpca/fetchcode.php?id=210?ExtendedYaleB.mat"

wget -O data/Hopkins155 "http://www.vision.jhu.edu/data/fetchdata.php?id=1?Hopkins155.zip"

You can now skip the following subsection to create symbolic links

If you already have downloaded or two of the datasets, you can create your own symbolic link, or use the corresponding tools/link_{extended_yale_b,hopkins155}.sh script to properly create a fresh link in the required data/ directory.

To use link_hopkins155.sh, just enter:

bash tools/link_hopkins155.sh

You can find all results in the report provided with the repository in report directory. Here are extracted two figures from the report/report.pdf file:

@book{vidal2016PCA, title={Generalized Principal Component Analysis}, author={Vidal, Ren{'e} and Ma, Yi and Sastry, S Shankar}, year={2016}, publisher={Springer} }