- Import your own existing
ModelVersionand associated training script
- Store it in your Private Registry
If you do not want to train your
ModelVersion with Picsellia or if you already have a
ModelVersion ready to be used and deployed, you can directly import & operate it with Picsellia.
ModelVersion will end up in your Private Registry as is the case for any other model you might have trained with Picsellia.
Click New Model from the Private Registry and define your
ModelVersion name and description to do it.
It is now time to create your first
ModelVersion, once again, click Create version and fill in the modal. At this step, you'll be asked to enter your
ModelVersion name, description, type, framework,
Labelmap, and training script parameters as shown below.
The last point is to fill in the Docker Image link of the training script previously adapted with Picselia SDK and dockerized (a user guide to help you dockerizing your _training script _is available here).
Labelmapand parameters remain editable but make sure info remains consistent to avoid further errors
Model and its first version are now created! To complete the upload procedure, you just need to store the model files by clicking on + like this:
In case you have the Datset used to train the
ModelVersion, you can upload it as a
DatasetVersioninto Picsellia (as explained here), it will be useful later to compute unsupervised metrics (such as datadrift detection) in the Deployment part of Picsellia
ModelVersion is now fully integrated on Picsellia and can be operated!
Updated 21 days ago