Deployments - Inferences

Until now, the different features related to a Deployment have been described. So you should now have a fully and properly configured Deployment. It is then, high time to log the result of inferences in it.

The ModelVersion that is related to the current Deployment can be deployed either:

• On the Picsellia Serving Engine
• On your own serving infrastructure

1. If the ModelVersion is deployed on Picsellia Serving Engine

If the ModelVersion is deployed on the Picsellia Serving Engine, you need to use the Picsellia Python SDK to make this ModelVersion make inferences on your image.

Here is the recipe to follow to make your ModelVersion deployed on the Picsellia Serving Engine doing an inference:

The fact that the ModelVersion is deployed on Picsellia Serving Engine is quite convenient as when the inference is done, the PredictedAsset and associated Prediction are automatically and properly logged by Picsellia on the associated Deployment.

You can then visualize the PredictedAsset and associated Prediction using the Predictions overview and perform Reviews using the Prediciton Review tool for instance.

2. If the ModelVersion is deployed on your own serving infrastructure

If the ModelVersion is deployed on your own serving infrastructure you are in charge of making your ModelVersion infer properly. In addition, you will have to adapt the script that is performing the inference on your infrastructure to make it log the result of the inference (i.e. the PredictedAsset and the associated Prediction).

Here is the procedure to apply to log the result of your inference on a dedicated Picsellia Deployment.

Once logged, you can then visualize the PredictedAsset and associated Prediction using the Predictions overview and perform Reviews using the Prediciton Review tool for instance.