Deploy Pipeline with Intel® Geti™ Platform Model

Follow this procedure to configure a pallet defect detection pipeline that uses a model trained with Intel® Geti™ platform.

Note
Learn more about training your own models for object detection, classification, segmentation, and anomaly with Intel® Geti™ platform.

1. Copy the model created with Intel® Geti™ platform into the following directory in the reference implementation.

     resources/models/geti/<model_use_case>/deployment/
   

   Note
   You can organize the directory structure for models for different use cases.

2. Add the path to the model in the udfs section of the evam_config.json file located in the helm chart.

     "udfs": {
        "udfloader": [
           {
              "name": "python.geti_udf.geti_udf",
              "type": "python",
              "device": "AUTO",
              "visualize": "true",
              "deployment": "./resources/models/geti/<model_use_case>/deployment",
              "metadata_converter": "null"
           }
        ]
     }
   

3. If the reference implementation is running, restart the reference implementation before you proceed to the next step.

     helm uninstall pdd-deploy -n apps
     helm install pdd-deploy . -n apps  --create-namespace
   

4. Start the pipeline with the following cURL command. This example starts an AI pipeline.

     curl http://localhost:30107/pipelines/user_defined_pipelines/pallet_defect_detection_1 -X POST -H 'Content-Type: application/json' -d '{
        "source": {
              "uri": "file:///home/pipeline-server/resources/videos/warehouse.avi",
              "type": "uri"
        },
        "parameters": {
              "udfloader": {
                 "udfs": [
                    {
                          "name": "python.geti_udf.geti_udf",
                          "type": "python",
                          "device": "CPU",
                          "visualize": "true",
                          "deployment": "./resources/models/geti/pallet_defect_detection/deployment",
                          "metadata_converter": "null"
                    }
                 ]
              }
        }
     }'