Cluster By Family

[DanielGrin31]

Use Batch input for the family classifier and optimize similarity calculation

[DanielGrin31]

~1. Need to optimize by using vector search instead of cosine similarity matrix over all embedding vectors.~ there can be non-similar people from the same family so it may not work

groups=[]
for face_embedding_to_check in all_embeddings:
  face_embedding_to_check = [...]
  similar_embeddings= self.emb_manager.search(...)
  # calculate is_family for all similar vectors

   # add to groups

2. retrain family classifier.
3. False positive tends to merge family clusters
4. use probabilities for dbscan