Right now the face UMAP uses the OpenFace descriptors. These do not have a "canonical order" because it is a list of face not a list of photos.
To run UMAP on faces with supervision, we need:
- a .npy file of 32x32 pixel images for each face
- a .npy of the openface descriptors
- a .npy of one hot vector of genders and ages
All of these files should have the same number of rows, and each row should correspond to the same face.
Right now the face UMAP uses the OpenFace descriptors. These do not have a "canonical order" because it is a list of face not a list of photos.
To run UMAP on faces with supervision, we need:
All of these files should have the same number of rows, and each row should correspond to the same face.