Matching found pets to lost pet reports
Your task
You run a national website for lost and found pets. An owner posts a lost dog or cat with photos, plus where and when it went missing. Someone who finds a stray posts its photo and where they found it. Volunteers compare photos by eye today, and many pets never get home.
About 600 lost reports and 400 found reports are posted a day. A report stays open for 60 days or until it is matched. Most pets turn up within 15 km of where they went missing, and almost all within 50 km.
The data team has a model that turns a pet photo into an embedding, a list of 256 numbers. Photos of the same animal give close embeddings, but so can photos of different animals that look alike.
When a report is posted, which open reports do you compare it with, how, and who is told about a likely match?
It covers
- Embeddings
- Candidate filtering
- Human review