Dataset - Set up the dataset

Now that our DatasetVersion has been created and populated with Asset, it is time to start dealing with Annotation.

1. DatasetVersion Detection Type and Label

Before annotating the Asset composing our DatasetVersion, we need to initialize it by defining the Detection Type and the Label so that Picsellia will be able in the future to ensure consistency at each step of the process.

For instance, make sure that the Detection Type of a DatasetVersion is the same as the one of the ModelVersion you might be retraining in the frame of an Experiment.

🚧

If you plan to import Annotation though an annotation file

Then, you can skip this initialization part as this one will be done automatically by Picsellia while reading the information (Detection Type and Label) contained in your annotation file during the import.

A. Set DatasetVersion Detection Type

To set properly your DatasetVersion, you must go to Settings of the DatasetVersion and jump to the Labels tab. From here, you can first select your Detection Type among Object Detection, Classification, Segmentation, Line, and Point and click on Set dataset version type.

🔄

Change DatasetVersion Detection Type

You can change the Detection Type of a DatasetVersion if already defined, but to do so, the DatasetVersion must be free of Label meaning deleting all existing Label.

B. Create Label

Once defined, you can start creating your Label, there are two ways to do it.

a. From Scratch

Thanks to the button + Add new label you will create a new and empty Label line. For each new line, you will be asked to give a name to this new Label and click on + to save it.

b. Importing them from another DatasetVersion

Thanks to the button Import label names from another dataset version you will be able to browse among the Label of other DatasetVersion of the current Dataset. So you just need to select the DatasetVersion name from which you want to import and select the Label names to import in the current DatasetVersion and click on the import icon.

Please note that as only the name of the Label is imported at this step, you can import Label names from DatasetVersion that has another Detection Type.

When all your Label are created, you can visualize them and the number of Shape in the current DatasetVersion with this Label. The set of Label attached to a DatasetVersion is called the Labelmap.

C. Edit Label

At any moment, you can change the name of a Label by clicking on its name. The name modification will be taken into account as soon as you click on the check mark icon. This operation will change the Label value of all existing Shape with this Label in the Annotation of the current DatasetVersion.

You can also decide to delete one a several Label from your DatasetVersion.

Before doing that you must be aware that deleting a Label will delete all the Shape with this Label from the Annotation in the current DatasetVersion. So use this feature with caution.

If you decide to delete all the Label of a DatasetVersion, it will delete all the Shape from all the Annotation in the current DatasetVersion, meaning that all Annotation will still exist but without any Shape.

2. What is an Annotation?

An Annotation is a set of Shape that are linked to an Asset in a DatasetVersion. Each Shape has a Label among the ones defined previously.

For instance, this Asset has an Annotation that contains 2 Car Shape and 3 Person Shape.

A DatasetVersion can be considered as usable for training when all the Asset have at least one Annotation.

There are three ways to create or modify Annotation on Picsellia:

• Import Annotation file
• Annotation tool
• Import Annotations from another Dataset Version

In summary, an Asset of a DatasetVersion has up to one associated Annotation. Each Annotation can have several Shape, each Shape has a Label among the Labelmap defined in the DatasetVersion Settings.

3. Visualize Annotation

Now let's consider our DatasetVersion as already annotated.

Thanks to the different views proposed in your DatasetVersion, you can see the Metadata related to Annotation in Table and Details views:

And visualize the different Shape displayed on the image of your Asset through the Grid and Details views:

👓

Personalization of Shape visualization

You can select the Shape you want to visualize on your Asset based on the Shape Label using the button x/x Labels in the top-right corner of the view (with x being the number of Label in the Labelmap of your DatasetVersion)

4. Delete Annotations

You can delete all the Annotation related to one or several Asset, by selecting them, and clicking on More > Delete annotations.

🐍

To delete one Annotation of an Asset that has many, you can use the Python SDK

5. Analytics

Once your DatasetVersion is annotated, Picsellia will compute several metrics that can be used to assess the quality of your Dataset.

To visualize them, you can jump into the Analytics tab of your DatasetVersion:

📘

In case an Annotation Campaign is ongoing on the current DatasetVersion

In case an Annotation Campaign is ongoing, the Analytics tab will be enriched with additional metrics, they are described in details here