Be able to read in facial landmark positions as input from Computer Vision group

[diegocasmo]

• [x] Parse Computer Vision input (see format)
• [x] Export/import best model (either MLP or SVM)
• [x] Use best model predict Computer Vision input

Notes:

• Kaylan mentioned in the meeting that we need to be able to tell what features are the most important ones and communicate that to the Computer Vision group. It would be a good idea to send an email to Kaylan to clarify this requirement.

[tristaaan]

So the input is an array of facial expressions?

We can figure out the principal components, but it will vary dataset to dataset.

[diegocasmo]

I'm also a bit confused about this, since PCA creates the principal components based on variance, which is dependent on the measurements :/

[tristaaan]

I think we have a good start on this in a branch, we need to resolve the accuracy drop (I imagine) we'll get when we take out PCA. Might sit on this a bit for Monday.

[diegocasmo]

I may be wrong, but I have been thinking about this, and maybe we won't need to take PCA out of the equation at all, instead we'll used it normally as we always do. PCA selects n components that have the largest possible variance (not n features!). So unless the input face is an outlier, this should be fine.

[tristaaan]

We need variance across a single column, one row gives no good results. Two rows not much better.

import numpy as np
from sklearn.decomposition import PCA
pca = PCA(n_components = 5, whiten = True)
principal_components = pca.fit_transform(np.array([[1,2,3,4,5,6,7,8,9,11]]))
principal_components

[diegocasmo]

This should help: https://github.com/scikit-learn/scikit-learn/blob/a24c8b464d094d2c468a16ea9f8bf8d42d949f84/sklearn/decomposition/base.py#L99. I think X is pca.components_.

Documentation: http://scikit-learn.org/stable/modules/generated/sklearn.decomposition.PCA.html#sklearn.decomposition.PCA.transform