LabelStudio tooling

Set of tools and scripts to leverage LabelStudio to import/export/check annotations made by CV models.

Workflows

Import and Review ultralytics YOLO predictions

1. Import the ultralytics dataset

uv run label-studio-converter import yolo \
  -i ./datasets/ultralytics/coco8/ \
  -o data/import/ls-tasks.json \
  --image-root-url "http://localhost:8000/ultralytics/coco8/images/"

2. Start the label-studio environment

docker compose up

3. Follow the instructions from the import command to import data into label studio

Export LabelStudio Annotations from LabelStudio in YOLO format

1. Select the Export button in the Project > YOLO with Images
2. All the label txt files have their updated bbox coords

Scaffolding

• The datasets are located in ./datasets and the folder will be served statically by the HTTP server.
• ./mydata is used by label-studio to store data locally.
• ./docker contains the Dockerfiles

YOLO Datasets

This is the coco8-import dataset:

.
├── classes.txt
├── images
│   ├── 000000000009.jpg
│   ├── 000000000025.jpg
│   ├── 000000000030.jpg
│   └── 000000000034.jpg
└── labels
    ├── 000000000009.txt
    ├── 000000000025.txt
    ├── 000000000030.txt
    └── 000000000034.txt

The classes.txt file should contain all classes in the right order:

person
bicycle
car
motorcycle
airplane
...
teddy bear
hair drier
toothbrush

Commands

Start label-studio and a local HTTP server to serve images from a dataset:

docker compose up

Note: Label Studio is started on port 8080 and the HTTP static server on port 8000.

Import a Ultralytics dataset:

uv run label-studio-converter import yolo \
  -i ./datasets/ultralytics/coco8/ \
  -o data/import/ls-tasks.json \
  --image-root-url "http://localhost:8000/ultralytics/coco8/images/"

Note: It will generate a ls-tasks.json and a ls-tasks.label_config.xml files that can be imported in a label studio project.

Terraform Deployment

Deploy LabelStudio to AWS using the provided Terraform configuration:

Prerequisites

• Terraform installed
• AWS CLI configured with appropriate credentials
• An AWS account with necessary permissions

Deployment Steps

1. Navigate to the terraform directory:

   cd terraform/

2. Copy and configure variables:

   cp terraform.template.tfVars terraform.tfVars

   Edit terraform.tfVars with your values:

   • s3_bucket_name: Unique S3 bucket name for storing LabelStudio data
   • db_username: PostgreSQL database username
   • db_password: Secure PostgreSQL database password
   • db_name: Database name (default: labelstudio)
   • labelstudio_username: Admin username for LabelStudio
   • labelstudio_password: Admin password for LabelStudio
   • labelstudio_token: Random token for API access
   • aws_region: AWS region (default: us-west-2)
   • instance_type: EC2 instance type (default: t3.small)

3. Initialize Terraform:

   terraform init

4. Plan the deployment:

   terraform plan -var-file="terraform.tfVars"

5. Apply the configuration:

   terraform apply -var-file="terraform.tfVars"

6. Access LabelStudio: After deployment, use the output instance_ip to access LabelStudio at http://<instance_ip>

Infrastructure Components

The Terraform configuration creates:

• EC2 Instance: Runs LabelStudio in Docker container
• RDS PostgreSQL: Database for LabelStudio data
• S3 Bucket: Storage for uploaded files and exports
• IAM Roles/Policies: Secure access between services
• Security Groups: Network access controls

Cleanup

To destroy the infrastructure:

terraform destroy -var-file="terraform.tfVars"

Resources

• LabelStudio Blog: Importing Local YOLO Pre-Annotated Images to Label Studio