This is a Face Recognition learning repo for research.
- Conduct parameter tuning.
- Log the params for the algorithms.
- Caching working for all cases, Yay!
- Conduct an ensemble classifier on all of the metric learning based techniques. Partly done, though hard voting is also required to be implemented.
- Port LFW: http://scikit-learn.org/stable/modules/generated/sklearn.datasets.fetch_lfw_people.html
- Test why the 40 img dataset gave a segfault: Done, this is because we fed a k > number of points per label.
- Do the preprocessing as conducted by them, and try out kNN with PCA on Euclidean and Mahalanobis distance, using
the
get_distance()API given by Shogun. - Get LMNN to work. Edit: Needs to be made ready for our metric learning algorithms.
- Modify the existing NN function to give a kNN, for better comparison.
- Enable LFDA, wrap it
- Mangle the data in the format required by the metric-learn module and feed it for results. Edit: Doing this inside the classful implementation itself.
- Managed to get LDML, LFDA, LMNN, LSML, RCA working with good accuracies.
- Try out Gabor Features, to try to analyse the use of metric learning algorithms in that space.
- ITML
- LMNN
- LSML
- SDML
- LDML
- NCA
- RCA