Dataset - Annotation import and export
Picsellia allows you to create your Annotation
by importing your own Annotation file on an existing DatasetVersion
.
1. Annotation formats
Picsellia natively handles the following Annotation formats:
- COCO
- Pascal VOC
- YOLO
- ViaLabel
More documentation about those formats is available here
2. Import Annotation files
From the Assets overview, you can click on Annotations > Import to open the Annotation files import modal.
From this modal you must indicate:
- The format of the Annotation file
- The expected behavior of the
DatasetVersion
already containsAnnotation
created by your user (keep existingAnnotation
, replace with newAnnotation
, concatenate new and existingAnnotation
) - The files containing
Annotation
related to the images composing the currentDatasetVersion
. - The behavior if you import a Segmentation Annotation file, knowing that Picsellia can import those
Annotation
as Polygons or _Bounding Boxe_s.
If the files contain Annotation
of images that are not part of the DatasetVersion, Picsellia will simply ignore them and only import the Annotation
related to images that are part of the current DatasetVersion
.
Then you just need to click on Import and the Annotation will be created on your DatasetVersion
. You can now visualize through the Assets overview the Annotation
(Shapes
and Metadata)
No need to setup
DatasetVersion
Detection Type andLabelmap
As the Detection Type and the
Labelmap
are contained in the Annotation files, they will be automatically setup on your PicselliaDatasetVersion
during the import. For sure, after the import you can still create manually additionalLabel
.
3. Edit Annotation
Annotation
Once imported the Annotation
can still be modified directly on Picsellia using the integrated Annotation Tool.
Annotation Tool documentation is available here.
4. Export Annotation
Annotation
At any moment you can export the Annotation
of all the Asset
of an annotated DatasetVersion
. You can choose the format of the Annotation
you are about to export among COCO, PascalVOC, YOLO, and ViaLabel.
To do so, you can click on Annotations and choose the Annotation format. The file will be generated by Picsellia and downloaded on your machine. The Annotations export process can take several seconds depending on the DatasetVersion
size so please be patient.
In case an Asset
has more than one Annotation
, it is the one displayed in the Details view that will be exported, meaning either the one in status Accepted or the most recently updated one.
Personalize the Annotation export
For more customization in the export of
Annotation
, you can use the Python SDK
5. Duplicate Annotation
Annotation
In case an Annotation
has to be reproduced on all the unannotated Asset
of a DatasetVersion
, you can use the Duplicate button.
To do it properly, you need to select the annotated Asset
of your DatasetVersion
, for which the Annotation
has to be duplicated and click on Duplicate. A modal will open, in this one, you just need to click on Duplicate as shown below:
Automatically, all the Asset
of the current DatasetVersion
that doesn't have Annotation
yet will get the Annotation
of the selected Asset
. It means that the existing Annotation
won't be overwritten.
Updated 10 months ago