QA Tool

This repository is for machine learning dataset quality assurance application.

It only supports Vision-Language dataset in current version.

Examples

View of data list. Check shows if you have inspected the data. You can move to the inspectation view by clicking the link in the detail.

By clicking the button accept and reject, you can send the result to the database. The button is hilighted unless you change the quality status.

Usage

Environments

• Debian-bookworm
• Python 3.12.1

Requirements

Docker-compose

If you haven't installed docker-compose, you could install it by the following script.

sudo apt-get update
sudo apt-get install docker-compose-plugin

Dataset Preparation

Images

You should place your images into /data/vlm/images

Text

You can upload the data from local(File size should be under 1GB).
The format of LLaVA is only supported for now.

Deploy

You could deploy the server by ruunning the following script.

docker-compose up --build -d

The annotation qa tool will be deployed at the following url if you are going to use it in local.

http://localhost:30010

Data Upload

You can upload the data by using the upload feature in the navigation bar.

Export Data

You can download the data by using the export feature in the navigation bar. The downloaded data will be saved in /data/{datasetname}/exported.

Database

To see the datas in the MongoDB can be seen using mongodb compass (Because express isn't running). The server url would be

mongodb://root:1234@localhost:27017

Note that this is the initial id and password. You could change the name and password in docker-compose file. After chaning the name and the password of mongodb you should change the DB_URL in the env file

You can use mongosh by

docker exec -it qadb mongosh

Security

For secure datas you need change the SECRET_KEY in the env file. Also the name and password of the mongodb

Serving as server

You need to change the ip address in nginx setting file and frontend env file

Future update

• Admin user
• User Profile
• Supporting other dataset types
• Adding Annotation Feature