In the past we used FaceNet to extract emebeddings from faces and then cluster them using the "chinese whipers" algoritm.
Unfortunately, it required substantial hand tuning to find the correct parameters..
It's possible using UMAP and HDBSCAN algorithsm (like this for document: https://towardsdatascience.com/topic-modeling-with-bert-779f7db187e6), would make this process more automatable, and possibly could implement as a single pliers Extractor.
In the past we used FaceNet to extract emebeddings from faces and then cluster them using the "chinese whipers" algoritm.
Unfortunately, it required substantial hand tuning to find the correct parameters..
It's possible using UMAP and HDBSCAN algorithsm (like this for document: https://towardsdatascience.com/topic-modeling-with-bert-779f7db187e6), would make this process more automatable, and possibly could implement as a single pliers
Extractor.