Getting started with Label Studio

Welcome to Label Studio! In this video we'll walk you through the ins and outs of installing Label Studio,

creating your first project, and getting started with some of the features that are available to you as an open source user.

At the time of recording, Label Studio offers three project tiers: Open Source, which is freely available to download and run under the Apache 2 license, Starter Cloud, which is hosted with a few additional features on a pay by month plan, and Enterprise, our most robust and secure labeling and evaluation tool which is paid on a yearly basis. To learn more about the differences between these products, visit us at www.humansignal.com/pricing.

Today, we'll be focusing on open source, our basic product, which is a great place to start if you're new to data labeling and human in the loop model evaluation.

Label Studio open source is run and hosted on your local machine. To do this, we can use a Python environment using Conda. First we'll create a new conda environment: conda create --name

we'll call it LSO for Label Studio open source and we'll use Python=3.12. We'll give it a minute to load.

We'll press y and we'll give it a minute to extract. Now, we'll activate it Conda activate LSO. Now that it's activated we can simply run pipinstall -U label-studio to install Label Studio to our local machine.

If you can't use pip check out our website for instructions on how to install using other methods.

Now that the tool's installed, all we have to do is run label-studio start to kick off the program. It's going to create our database for us, set our secret keys and launch the program on localhost:8080 if it's available or the next open port if it's not.The first time you run the program it takes a minute to kick off, but usually it's much faster.

You can see here that by default Label Studio runs on localhost:8080 if it's available or the next available port if it's not. This is the URL that we'll use to access Label Studio from our local machine. If you want to deploy Label Studio open source to the cloud or an internal server, or have folks who are not on your local machine access it,  make sure that the port that you use is accessible to the outside world or consider using a tunneling system like ngrok. You can learn more about these issues in our documentation.

Now that Label Studio is up and running, we need to make an account. Click on the blue sign up button here in the bottom corner, provide your email address and create a password. Then you'll be logged in. I'm gonna go ahead and do that and I'll be back.

Now that we're logged in, you're greeted by the welcome screen, which has a few quick start buttons here at the top, as well as an area where your recent projects will populate if you have any, and some helpfu resources here on the right hand side. Since we don't have any projects right now nothing populates. Let's start by creating our first project by clicking the Create Project button. We'll give it a name, here, we'll call it getting started demo, and there's a place for us to add a description if we want, which is great for extra project level information you might want to keep track of like data versions. For now we'll click on the Data Import tab.

Label Studio accepts a wide variety of data types ranging from single entity files like JPEG and MP4

to dataset quantity files like CSV and JSON and everything in between. There are a few different ways to upload your data. You can drag and drop individual files (single entities or datasets) into the blue box to import, or you can click on the Upload Files button here, to access your file storage locally or you can use a URL to an acceptable data type to upload. Here , I have the URL to a JSON file that contains audio, video and image data. I'll go ahead and paste that link into the Dataset URL box and click add URL. This will upload every item in my JSON as a new task in Label Studio. You can also connect your Label Studio instance to local or cloud storage from the storage tab in your project settings. You can learn more about external and local storage options in the documentation. For now we'll click the Labeling Setup tab at the top of the screen.

Label Studio's bread and butter is the ability to create totally customizable UIs for labeling pretty much any kind of data you might need. To do that, we use a custom XML based system to define each element of the task UI. On this screen, you'll see a collection of templates to help you get started with some of our most common use cases. Simply find one that meets your needs and select it to get started. Here, we're going to use the Bounding Box Detection project. The screen that you'll see is our visual editor, a low code way of interacting with your labeling config, or the file that defines the UI for labeling a project in Label Studio. If you want to interact with the code directly, click on the code tab shown here. This screen allows you to edit the XML directly, allowing you to completely customize the look and feel of your system. You can combine as many data and label types as you may need but remember: the more complicated your project, the harder it will be to get quality output from your annotators. For the purposes of this demonstration we're going to leave the image classification as is with one small change: we'll change the options "airplane" and "car" to "dog" and "cat". For a deeper understanding of the power of labeling configs, check out the tag section of our documentation. Now we'll click on the blue save button in the upper right hand corner of our screen to save and open our project.

Congratulations you've created your first project! Now you're ready to label. This screen is the Data Manager, and it's the screen you'll see when you click into any existing project in Label Studio. Each row in the Data Manager represents a single task, or item to label, in your project. On the left side of the Data Manager you'll see some columns that correspond to Label Studio metadata including an internal ID number and statistics about how many people have labeled or skipped or pre-annotated the task, as well as who annotated it. After that you'll see all of the data that you've uploaded. In our case, we have image, audio, and text. We strongly recommend that if you have an internal ID number for each task that you upload that as metadata within your task so that you can resolve the information more easily later. Now, all we have to do is click on the label all tasks button shown here to begin labeling. This button will bring you to the next unlabeled task in your project. If you want to label a specific task, you can simply click into it from the Data Manager. When we click on the label all tasks button, you'll see that our next unlabelled task appears. In this case, we see that we have an image with the labels at the bottom. Since all we're doing is drawing bounding boxes, we can click on the dog label and draw a box around the dog that we see and then click submit. And that's it! We've begun labeling our data.  We can continue in this manner throughout every task we have in our project.

There are a ton of different settings that you can configure for each project that you make in Label Studio. Click on the settings button here at the top to see them. In General Settings, we can change the name and description or color of our project.  In Labeling Interface, we can change the interface as we go if we learn more about ways to make it easier on our annotators. Annotation Settings allow us to add labeling instructions or pre labeling instructions. The model settings is where you'll go to hook up a ML backend, the way that we connect any model that you might have to Label Studio for the purposes of pre labeling or annotation. To learn more about the ML backend, check out our other documentation. We can also change Prediction Settings, Cloud Storage,  Webhooks,  and more!

Now that you have a project it's unlikely that you'll be labelling alone. Let's see how to add new members to a Label Studio instance. Click on the hamburger menu up here, next to the Label Studio logo in the upper left hand corner, and select the Organization tab. This brings you to the Organization page, where you can manage your members. Click on the blue add people button shown here and copy the link shown and send it off to whomever you need to label. They'll create their own account and be added to your instance. Looking for more control over who can see what data and take particular actions? Check out Enterprise and Starter Cloud editions that offer role based access control.

Today you learned how to get started with Label Studio Open Source but, there are a ton of features we weren't able to cover. Check out our other videos and the documentation linked in the video description to learn more You can also join us on our Community Slack page to connect with other users and get help on your Label Studio journey.

Happy labeling!