Skip to main content

animals & objects

silo uses object detection to identify animals, objects, and scenes in your photos, making it easy to find specific content.

object detection​

yolo (you only look once) models identify various objects in your images:

  • animals (dogs, cats, birds, etc.)
  • vehicles (cars, bicycles, boats)
  • household items (chairs, laptops, cups)
  • outdoor objects (trees, buildings, signs)

animals section​

navigate to the "animals" tab to see detected animal clusters:

  • automatically grouped by species
  • displays sample images
  • shows total count per type

common animals detected​

  • dogs (various breeds)
  • cats
  • birds
  • horses
  • wildlife (deer, bears, etc.)

naming animal clusters​

similar to face clusters, you can name specific animals:

  1. click on an animal cluster
  2. enter your pet's name (e.g., "spot", "whiskers")
  3. click "save"

now you can search for your pet by name:

  • "spot playing fetch"
  • "whiskers on the couch"
  • "spot and blake at the park"

known issues​

caution

the animals section is still in development and has some limitations.

current issues:

  • some inanimate objects are incorrectly classified as animals
  • clustering accuracy varies by species
  • rare animals may not be detected

we're working on:

  • improved classification accuracy
  • better cluster separation
  • expanded species recognition
  • manual correction tools

even without dedicated tabs, you can search for objects semantically:

  • "photos with laptop"
  • "bicycle in garage"
  • "coffee mug on desk"
  • "red car"

scene detection​

silo can also identify scenes and environments:

  • indoor vs outdoor
  • natural vs urban
  • day vs night
  • weather conditions

example searches​

  • "outdoor sunset"
  • "indoor party"
  • "snowy mountain"
  • "rainy day"

improving accuracy​

better detection​

help improve object detection:

  • use well-lit photos
  • avoid extreme angles
  • ensure objects are clearly visible
  • higher resolution images work better

false positives​

if you see incorrect detections:

  • they won't affect semantic search accuracy
  • clip embeddings still capture true content
  • manual corrections coming in future update

future features​

planned improvements:

  • manual object tagging
  • custom object categories
  • training on your specific objects
  • improved animal classification
  • separate tabs for vehicles, plants, etc.

settings​

adjust object detection settings:

  • confidence threshold
  • object size minimum
  • categories to detect
  • enable/disable specific classes