---
updatedAt: 2026-09-03T15:45:37.000Z
agentTools:
projectIndex: https://documentation.picsellia.com/llms.txt
---
# Attributes
An `Attribute` is a structured field that lets you capture information beyond a `Shape` or an entire `Asset`: a color, a severity level, a quality score, or any other property specific to your use case. Two kinds of `Attribute` coexist on Picsellia: `Shape Attribute`, attached to a `Label` and filled in per `Shape`, and `Asset Attribute`, attached to the whole `Asset`.
For Computer Vision teams, `Attribute` are what turn a `DatasetVersion` into more than a set of `Label`: they let you run structured quality control, filter or aggregate on criteria that a `Label` alone can't express, and capture business-specific information (severity, quality grade, condition) directly next to the `Shape` or `Asset` it describes, instead of relying on free-text notes or spreadsheets kept outside Picsellia.
## 1. `Shape Attribute`
A `Shape Attribute` is defined at the `Label` level and is filled in when annotating for each individual `Shape` using that `Label`.
It is created and defined from the `DatasetVersion` *Settings*, in the *Labels* page, using the *+ Add attribute* button next to a `Label`, as introduced [here](https://documentation.picsellia.com/docs/dataset-set-up-the-dataset#d-add-attribute-to-a-label).
Creating or editing a `Shape Attribute` lets you configure:
* A *name*.
* A *Type*: *String*, *Integer*, *Float*, or *Boolean*.
* Whether it is *Required*.
* Under *Advanced parameters*: a predefined list of *Choices* (not available for the *Boolean* type), *Allow multiple answers* to let several *Choices* be selected at once, and, for *Integer* or *Float*, a *Min* / *Max* bound.
Once all the `Shape Attributes` for the selected `Label` are properly defined, you can click on the *Create* button and potentially do the same for any other `Label` of the current `DatasetVersion` that needs `Shape Attributes`. The number of `Shape Attributes` defined per `Label` is then displayed in the *Labels* page as shown below:
You can edit or delete an existing `Shape Attribute` at any time, by clicking on the *...* button next to any `Label`.
The *Delete Attributes* option will permanently delete all the `Shape Attributes` configuration and all the associated values for the selected `Label`.
By clicking on *Update Attributes*, you will have the possibility to edit the configuration (Name, Type, whether it is required, the Advanced Parameters) for each `Shape Attribute` of the selected `Label`. Once the changes are done, you simply need to click *Update* to take them into account. Please note that it won't affect the values already created in the *Annotation Tool* (or via SDK), only the to-be-created ones will be guided by the newly edited configuration.
From the *Shape Attributes* update form, you can also perform a soft delete of any `Shape Attributes`; rather than being completely deleted, a `Shape Attribute` can also be *archived*: it disappears from the annotation experience in the future while its already-recorded values are preserved, and an archived `Attribute` can no longer be *Required*. In that case, you need to click on *Delete config only*. If you want to delete the `Attribute` config and its associated value, click on *Delete config and values*.
## 2. `Asset Attribute`
An `Asset Attribute` uses the same configuration mechanism as a `Shape Attribute` (name, *Type*, *Required*, *Choices*, *Allow multiple answers*, *Min* / *Max*), but is attached directly to the `Asset` as a whole rather than to a single `Shape`. This lets you capture information that describes the entire `Asset` rather than one specific object within it, for instance the weather conditions or the acquisition device for an image, independently of how many `Shape` it ends up being annotated with.
Unlike a `Shape Attribute`, an `Asset Attribute` isn't attached to any particular `Label`. It is configured from the same `DatasetVersion` *Settings* *Labels* page, but through the *...* menu in the header of the `Label` table (next to the `Label` count), rather than through a specific `Label` row.
The `Asset Attribute` creation form works the same way as `Shape Attributes`:
Once created, it also works the exact same way as `Shape Attributes` to update the configuration or delete `Asset Attributes`, whether it is a complete deletion or an archiving of the values.
## 3. Fill in `Attribute` values
We now consider that all the necessary `Asset Attributes` and `Shape Attributes` have been defined as detailed in the previous paragraphs. It is time to check how to give values for these, for each `Asset` or `Shape`.
`Attribute` values are mainly filled in from the *Annotation tool* while annotating. In the Annotation Tool, you need to access the Selection tab accessible from the right-side floating menu as shown below:
Here, either none or several `Shape` are selected; in that case, the `Asset Attributes` will be displayed, with the possibility for the user to fill in the available `Asset Attributes`.
Or only one `Shape` is selected; in that case, the *Selection* menu will display all the information that is related to this one, meaning `Label`, *Text* field, coordinates (for Bounding Boxes and Keypoints only), and the `Shape Attributes`. From that Selection menu you'll be able to fill in the different values (as defined in the configuration) for all the `Shape Attributes` that are related to the `Label` of the selected `Shape`.
### Required Attributes
If a `Shape` or an `Asset` is missing a value for a `Required` `Attribute`, a warning icon will be highlighted and the `Annotation` cannot be saved in the _Annotation tool_ until every Required `Attribute` is filled in.
## 4. Manipulation of `Attribute` (Annotation Import/Export, SDK, Fork..)
* When [forking](https://documentation.picsellia.com/docs/dataset-fork-a-dataset) a `DatasetVersion`, the existing `Attribute` configurations (`Asset` and `Shape`) and the associated values for the forked objects are kept.
* When [importing `Label` from another `DatasetVersion`](https://documentation.picsellia.com/docs/dataset-set-up-the-dataset#b-importing-them-from-another-datasetversion), each `Label`'s `Shape Attribute` configuration can be carried over along with it if the associated checkbox is ticked.
* Exporting a `DatasetVersion` to a *COCO* annotation file includes an `attributes` key filled with the `Shape Attribute` and `Asset Attribute` values of your `Annotation`. Importing a *COCO* annotation file that contains such an `attributes` key saves those values back onto Picsellia.
* On video `Asset`, `Attribute` values are saved on the `Track` of an `Annotation`. They are not yet included in *COCO* exports.
* From the SDK, the `Attribute` configuration is managed with `get_attributes()` / `set_attributes()`, available on both `Label` (`Shape Attribute`) and `DatasetVersion` (`Asset Attribute`). `Attribute` values themselves are exposed through the `attributes` property, available on an `Annotation` (`Asset Attribute` values) and on every `Shape` type (`Shape Attribute` values).
## 5. Filter on `Attribute` values
Recording an `Attribute` value pays off when you can retrieve the `Asset` carrying it: isolate every `Asset` graded below a quality threshold before a review round, gather the `Shape` marked as occluded to check how the model behaves on them, or count how many `Asset` were captured in a given condition before forking a `DatasetVersion` out of them.
`Attribute` are not searched through the *Query Language*. Instead, from the *Assets overview* of a `DatasetVersion`, use the *Add filter* button next to the *Search Bar*, detailed [here](https://documentation.picsellia.com/docs/dataset-overview#3-exploration-modes-and-filtering), and choose what to filter on:
* An `Asset Attribute`, matched against the value carried by the `Asset` itself.
* A `Shape Attribute`, matched against the value carried by a `Shape` of its `Annotation`. Since a `Shape Attribute` is defined on a `Label`, you select the `Label` first, then one of the `Attribute` configured on it. An `Asset` matches as soon as one of its `Shape` satisfies the filter.
The comparison proposed depends on the `Attribute` *Type*: an exact value for *String* and *Boolean*, with the configured *Choices* offered as suggestions when there are any, and a range for *Integer* and *Float*.
Lets start by creating a first filter on a `Shape Attribute` that is with the *Type String *and for which values can only be *Red* or *Green* as defined in the configuration by the `Attribute` creator.
Lets now add a second Attribute filter this time on an Asset Attribute that is an Integer:
Please note that the filters (whether they are configured on `Asset Attributes`, `Shapes Attributes` or *Custom Metadata*) can be used with the *SearchBar* to create advanced ways of filtering among your `DatasetVersion` .
Filters are also available on `Deployment` and `Experiment`
These filters are also available on the `PredictedAsset` of a `Deployment` and on the `Evaluation` of an `Experiment`, where they apply to the predicted and reviewed `Shape`.