Servant-Image-Server

Introduction

Servant-Image-Server (SIS) is a tool for labeling image data via object detection.

Implementation

Below are a some of the notable libraries used to implement SIS:

• servant libraries to create the web API/Server/Client.
• aeson for parsing JSON returned from API calls.
• sqlite-simple for all things persistence via sqlite3.
• wai/warp for actually running the webserver.
• hspec for testing.

Persistence

This tool utilizes sqlite3 in order to persistently store image and related data.

You will need access to sqlite3 in order for this tool to work.

Imagga

This tool is designed around interacting with Imagga, an online object detection API. As a result, a custom JSON parser was implemented around the expected JSON response format returned from calls to Imagga, and custom http requests were created in order to query its API.

In order to interact with Imagga's object detection API, you need to create an account. Once you do so, you will be supplied with your very own api key and api secret.

Note

This tools works with image URLs only.

Example Usage

SIS is easy to use.

Below is how to start up SIS:

$ cd server

$ stack build

$ stack exec server-exe [api-key] [api-secret]

Once the server has been started, you can begin to play!

In another terminal, you could do the following:

$ curl -X POST -d '{"imageinput_url":"https://docs.imagga.com/static/images/docs/sample/japan-605234_1280.jpg", "imageinput_label": "japan", "imageinput_enableobjectdetection": true}' -H 'Accept: application/json' -H 'Content-type: application/json' http://localhost:8080/images

• You can optionally specify "imageinput_enableobjectdetection": false should you want to skip querying and labeling via Imagga's API and store it in the backend.
• Additionally, you can omit the "imageinput_label" key and the server will label the image with the identifier that is generated for it.

This will query the API with the image found at https://docs.imagga.com/static/images/docs/sample/japan-605234_1280.jpg.

The response received will look something like below:

{"imageinput'_data":"...","imageinput'_enableobjectdetection":true,"imageinput'_identifier":10,"imageinput'_label":"japan","imageinput'_responsestatuscode":200,"imageinput'_url":"https://docs.imagga.com/static/images/docs/sample/japan-605234_1280.jpg"}

This will result in the server doing a few things:

• Downloading and decoding the image file in order to return the raw image data.
• Creating a sqlite identifier.
• Creating a label (if the user did not provide one).
• Inserting the identifier, label, and url into sqlite.
• Inserting rows for all labels (in another table different from the one above) returned from Imagga with a 50% confidence or greater.
• Returning the above plus the raw data and a boolean describing whether or not object detection was requested on the image.

You can also query the backend via a couple of different endpoints:

• $ curl http://localhost:8080/images
  • Returns metadata for all images stored in the backend.
• $ curl http://localhost:8080/images?objects="querystring"
  • The querystring has the format "string1,string2,...,stringn".
  • Returns image metadata for all images that contain labels as identified by Imagga.
• $ curl http://localhost:8080/images/id
  • Returns image metadata for the image stored at the id.
  • imageinput'_identifier in the response is the id to plug in here.

Test Suite

A simple test suite was implemented to ensure accuracy to the API specification.

You can run the test suite with the following commands:

$ cd server

$ stack test --test-arguments='[api-key] [api-secret]'