People Search

web service for finding similar people faces in Scala.

Introduction

Face recognition is one of the main problems of modern computer vision. There are many solutions developed for finding people from the pictures and authenticating them. Different governments and commercial companies use deep learning solutions, which in its core embed visual features of the person from an image into a vector of numbers.

This repository helps to build an API for such machine learning systems by providing database CRUD operations on the people face embeddings and clean endpoints for data manipulation.

Scala has been chosen as it induces pure functional style.

Stack used

Doobie - connection with database
PostgreSQL - database for storing embeddings
Http4s - manage requests and routes
Flyway - migrations
TypeSafe Config - reading config file
Circe - json encoding and decoding

How to start

To start the project run following commands:

# copy repository
git clone https://github.com/QazyBi/People-Search
# enter the repository folder
cd people-search/
# create container with sbt and code
docker build . -t sbt
# start services: web-server and postgresql database
docker-compose up -d  # -d flag refers to detached mode, remove if not needed

How to use

To get list of people in the database use next command:

curl [http://localhost:8080/people](http://localhost:8080/people)

To get information about specific person using he's id use:

curl http://localhost:8080/people/{id}  # specify id, for instance 1

To search top 5 similar people from the database to the person you specify

curl http://localhost:8080/search/{id}  # specify id of queried person, for instance 1

# one possible result
# [{"id":1203,"name":"Natalie Portman","embedding":[0.31,0.19,0.41,0.76,0.35, 0.3]},
#  {"id":562,"name":"Keira Knightley","embedding":[0.23,0.21,0.33,0.81,0.29, 0.14]},
#  {"id":1203,"name":"Britney Spears","embedding":[0.3,0.1,0.35,0.78,0.95, 0.04]},
#  {"id":1203,"name":"Kate Winslet","embedding":[0.12,0.23,0.45,0.70,0.89, 0.12]},
#  {"id":1203,"name":"Daisy Ridley","embedding":[0.45,0.16,0.24,0.84,0.91, 0.05]}]

To add new record of a person use following command:

curl -X POST --header "Content-Type: application/json" --data '{"name": "Martin Odersky", "embedding": [0.3, 0.1, 0.35, 0.85, 0.99, 0.01]}' http://localhost:8080/people

To update information about person in the database run following:

curl -X PUT --header "Content-Type: application/json" --data '{"name": "Martin Odersky", "embedding": [0.3, 0.1, 0.35, 0.85, 0.99]}' http://localhost:8080/people

To delete information about person in the database run following:

curl -X DELETE --header "Content-Type: application/json" http://localhost:8080/people/{id}  # specify id, for instance 1

Project Structure

resources:

db/migration/V1__create_people.sql - migrations that will be run when the project initializes

CREATE TABLE people
(
    id        serial       NOT NULL PRIMARY KEY,
    name     VARCHAR(100)  NOT NULL,
    embedding float8[]
);

SQL code creates a table with id, name and embedding fields.

Embedding is a vector of floats representing person in some latent space.

application.conf - configurations of the web server and of the connection to the database

scala:

config - ****module for reading configuration files(ServerConfig, DatabaseConfig)

db - module for database initialization through migrations and producing transactors

model - definition of the class mapping fields from PostgreSQL table

case class Person(
      id: Option[Long],  // corresponds to serial
      name: String,  // corresponds to VARCHAR
      embedding: List[Double]  // corresponds to SQL float8[]
)
case object PersonNotFoundError  // for handling cases when 
																// person not found in the table

repository - ****definition of methods for interaction with database.

getPeople - run sql query for selecting all people from the people table
getPerson - run sql query for selecting person matching with given Person id
createPerson - run sql query for creating a record in the table given Person class instance
deletePerson - run sql query for deleting record in the table given Person id
updatePerson - run sql query for updating record in the table given Person class instance

service - definition of routes and business logic(in further updates will be separated)

CosineSimilarity - object which defines useful methods for calculating cosine similarity between two lists of doubles.

Cosine similarity is a good measure of similarity of two vectors. Actively used when comparing two embeddings in machine learning.
PeopleService - defines routes for http requests and data manipulation.

HttpServer.scala - initialization of the web server

ServerApp.scala - starting point of the project

Tests reside in folders: src/it/scala/PeopleServiceSpec.scala, src/test/scala/service/PeopleServiceSpec.scala

What is next?

Adding image post requests
Adding stream processing

QazyBi / People-Search