How do we know that this principle components actually correspond to "Male" shape? Is it possible that other information is encoded in this component as well?
jacklishufan
In general, we do not because PCA is an unsupervised algorithm and is purely dependent on the data structure. The more precise way of saying this is that this principal component "aligns" well with the perception of "male". If you want to find a better representation of the "male" direction you can use methods such as LDA. Recently analogous ideas are also used in deep generative models to study latent representation of images.
LeslieTrue
Just as some additional information. Generally, each principal component usually represents a mixture of manually described features i.e. body shape, height, sexual features, etc. Hence, tuning any parameters of principal components won't directly create what you want unlike designing virtual characters in a 3D video game.
modatberkeley
What is the motivation behind finding the principal components of the male shape when in reality artists create these models?
Veriny
To add on to jacklishufan, the goal of PCA isn't to make a classification of "male" or "female", but rather to find patterns in the structure, in this case a blend shape.
How do we know that this principle components actually correspond to "Male" shape? Is it possible that other information is encoded in this component as well?
In general, we do not because PCA is an unsupervised algorithm and is purely dependent on the data structure. The more precise way of saying this is that this principal component "aligns" well with the perception of "male". If you want to find a better representation of the "male" direction you can use methods such as LDA. Recently analogous ideas are also used in deep generative models to study latent representation of images.
Just as some additional information. Generally, each principal component usually represents a mixture of manually described features i.e. body shape, height, sexual features, etc. Hence, tuning any parameters of principal components won't directly create what you want unlike designing virtual characters in a 3D video game.
What is the motivation behind finding the principal components of the male shape when in reality artists create these models?
To add on to jacklishufan, the goal of PCA isn't to make a classification of "male" or "female", but rather to find patterns in the structure, in this case a blend shape.