Skip to main content

face recognition

silo automatically detects and clusters faces in your photos, making it easy to find pictures of specific people without manual tagging.

how it works​

silo uses a multi-step process for face recognition:

1. face detection​

yolo and deepface models scan your photos for faces:

  • detects multiple faces per image
  • works with various angles and lighting
  • handles partial faces
  • processes videos frame-by-frame

2. face embedding​

each detected face gets a unique biometric signature:

  • 128-dimensional embedding vector
  • captures facial features mathematically
  • invariant to lighting and angle changes
  • enables similarity comparison

3. clustering​

hdbscan automatically groups similar faces:

  • no need to specify number of people
  • creates clusters based on similarity
  • handles noise and outliers
  • adapts to your photo collection

viewing face clusters​

navigate to the "people" tab to see all detected face clusters:

  • clusters are displayed as cards
  • each card shows representative faces
  • click a card to view all photos in that cluster

people cluster card

cluster details​

each cluster card shows:

  • sample faces from the cluster
  • total number of photos
  • cluster id (until you name it)
  • confidence score

naming clusters​

add names to clusters for easier searching:

  1. click on a cluster card
  2. enter the person's name
  3. click "save"

now you can:

  • search for that person by name
  • see their name in search results
  • combine with other search terms

merging clusters​

if the same person appears in multiple clusters:

  1. select the first cluster
  2. click "merge with..."
  3. select the second cluster
  4. confirm merge

all photos will now be in a single cluster.

splitting clusters​

if multiple people are in one cluster:

info

automatic split detection is coming in a future update. currently, you can manually exclude faces.

  1. open the cluster
  2. review all faces
  3. mark incorrect faces
  4. click "create new cluster from selection"

privacy features​

face recognition is completely private:

  • all processing happens locally
  • no face data sent to cloud services
  • face embeddings stored in local database
  • you can delete clusters anytime

managing face data​

view face database​

go to settings → manage database to see:

  • total faces detected
  • number of clusters
  • database size
  • last update time

delete face data​

to remove all face data:

  1. go to settings → manage database
  2. click "clear face data"
  3. confirm deletion

this removes:

  • all face embeddings
  • all clusters
  • face search indexes

your original photos remain untouched.

accuracy tips​

improve face detection​

  • use high-resolution photos
  • ensure faces are well-lit
  • avoid heavily filtered images
  • include multiple photos per person

clustering accuracy​

the clustering algorithm improves with:

  • more photos per person
  • variety of angles and expressions
  • consistent lighting
  • clear, unobstructed faces

known limitations​

face recognition may struggle with:

  • very small faces (distant shots)
  • extreme angles or occlusion
  • heavy makeup or costumes
  • significant age differences
  • very similar-looking people

settings​

adjust face recognition settings:

  • minimum face size
  • detection confidence threshold
  • clustering sensitivity
  • video frame sampling rate