This repository is for machine learning dataset quality assurance application.
It only supports Vision-Language dataset in current version.
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.
- Debian-bookworm
- Python 3.12.1
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
You should place your images into /data/vlm/images
You can upload the data from local(File size should be under 1GB).
The format of LLaVA is only supported for now.
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
You can upload the data by using the upload feature in the navigation bar.
You can download the data by using the export feature in the navigation bar. The downloaded data will be saved in /data/{datasetname}/exported
.
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
For secure datas you need change the SECRET_KEY
in the env file. Also the name
and password
of the mongodb
You need to change the ip address in nginx setting file and frontend env file
- Admin user
- User Profile
- Supporting other dataset types
- Adding Annotation Feature