image search engine

image-search-engine

What is it?

It is a search similar engine for images.

This is a web application allows you to search similar images in database.

The basic idea is that you can delete or add images at any time while maintaining data consistency. A useful feature is also implemented: search for similar images by request (for example, find similar images, but search only among those who are Labradors)

I am currently using:

• img2vec_pytorch - wrapper around Alexnet for image feature extraction (https://github.com/christiansafka/img2vec)
• postgeSQL with CUBE extension; vectors are very large, so i can't build index, but postgres allows to query data even in json fields

In last version i have used hnswlib it was faster, but not so flexible as postgres (you can check it out on 'hnsw' branch)

Deployment

Docker

docker-compose build
docker-compose up

Without Docker

requirements: postgeSQL; It will be simpler to run postgres in docker '/postgres/Dockerfile', otherwise you have to recompile CUBE extension (like in Dockerfile) also specify PG_USER, PG_DATABASE, PG_PASSWORD params in settings/env file

virtualenv venv --python=python3.6
source venv/bin/activate 
pip install -r requirements.txt
uvicorn run:app --host 0.0.0.0 --port 8001

App will be available on 0.0.0.0:8001 in both cases

Api Description

All handlers are available on 0.0.0.0:8001/docs

Add Image

POST /image add image to database

Python requests

import requests

r = requests.post(
    url='http://0.0.0.0:8001/image',
    files={'image': open('image_path', 'rb')}
)

Curl

curl -X POST "http://0.0.0.0:8001/image"
 -H "Content-Type: multipart/form-data" -F "image=@{image_path};type=image/jpeg"

Get Image

GET /image/{id} download image by id

Python requests

import requests

r = requests.get(url='http://0.0.0.0:8001/image/{id}')

with open('output_file_name', 'wb') as f:
    f.write(r.content)

Curl

curl -X GET "http://0.0.0.0:8001/image/{id}" --output {output_file_name}

Delete Image

DELETE /image/{id} delete image by id

Python requests

import requests

r = requests.delete(url='http://0.0.0.0:8001/image/{id}')

Curl

curl -X DELETE "http://0.0.0.0:8001/image/{id}"

Search Image

POST /image/search?k={k} search k nearest images

Most complex handler. You can search nearest images n all database or you can select only specific images (for example only 'Irish terriers') For such selects you need to add data to images as json fields (see POST data/{id}) Also you can select images by 'name' or 'path' in the same way. For such queries pass valid dict in params

Python requests

import json

import requests

r = requests.post(
    url='http://0.0.0.0:8001/image/search',
    files={
        'image': open('image_path', 'rb'),
    },
    params={'k': 3, 'query': json.dumps({'dog_type': 'Irish_terrier'})}
)

Curl

curl -X POST "http://0.0.0.0:8001/image/search?k={k}&query=%7B%22dog_type%22%3A%20%22Irish_terrier%22%7D"
 -H  "accept: application/json"" -H "Content-Type: multipart/form-data" -F "image=@{image_path};type=image/jpeg"

Add Data

POST /data/{id} add additional info for image by id

Pass all image data in json field

Python requests

import requests

r = requests.post(
    url='http://0.0.0.0:8001/data/{id}',
    json={'dog_type': 'Irish_terrier'}
)

Curl

curl -X POST "http://0.0.0.0:8001/data/{id}"
 -H "Content-Type: application/json" -d "{"dog_type":"Irish_terrier"}"

Get Data

GET /data/{id} get data for image by id (vector and some additional info)

Python requests

import requests

r = requests.get(url='http://0.0.0.0:8001/data/{id}')

Curl

curl -X GET "http://0.0.0.0:8001/data/{id}"

Query Data

POST /data/query get data for image by query

You can search for images by querying data (see POST /image/search) But you need to pass query data in json field

Python requests

import requests

r = requests.post(
    url='http://0.0.0.0:8001/data/query',
    json={'dog_type': 'Irish_terrier'}
)

Curl

curl -X POST "http://0.0.0.0:8001/data/query" -H  "accept: application/json"
 -H  "Content-Type: application/json" -d "{"dog_type":"Irish_terrier"}"