geti_deployment

Deployment of models trained with Intel® Geti™ Platform (gst-launch command line)

This set of samples demonstrates how to deploy models trained with Intel® Geti™ Platform.

How It Works

The Intel® Geti™ Platform defines a set of media analytics pipelines corresponding to common usage scenarios: classification, detection, segmentation, etc. In all cases, the platform outputs AI models in Intel® OpenVINO™ format: 'openvino.xml' and 'openvino.bin'.

This sample assumes a user has already trained models using Intel® Geti™ Platform and stored the output models in the followign folder structure:

-intel
 |-geti-anomaly-detection/          # model trained with Geti™ 'Anomaly detection PADIM/STFPM/UFlow' project
   |-padim/FP16                     # PADIM model
     |-openvino.xml                 # model metadata
     |-openvino.bin                 # model weights
   |-stfpm/FP16                     # STFPM model
     |-openvino.xml                 # model metadata
     |-openvino.bin                 # model weights     
   |-uflow/FP32                     # UFlow model
     |-openvino.xml                 # model metadata
     |-openvino.bin                 # model weights
 |-geti-detection/FP16              # model trained with Geti™ 'Detection bounding box' project
   |-openvino.xml                   # model metadata
   |-openvino.bin                   # model weights
 |-geti-classification-single/FP16  # model trained with Geti™ 'Classification single label' project
   |-openvino.xml                   # model metadata
   |-openvino.bin                   # model weights
 |-geti-classification-multi/FP16   # model trained with Geti™ 'Classification multi label' project
   |-openvino.xml                   # model metadata
   |-openvino.bin                   # model weights
 |-geti-classification-multi/FP32   # model trained with Geti™ 'Classification multi label' project
   |-openvino.xml                   # model metadata
   |-openvino.bin                   # model weights
 |-geti-obb/FP16                    # model trained with Geti™ 'Oriented Bounding Box Detection' project
   |-openvino.xml                   # model metadata
   |-openvino.bin                   # model weights
 |-geti-segmentation/FP16           # model trained with Geti™ 'Instance Seggmentation' project
   |-openvino.xml                   # model metadata
   |-openvino.bin                   # model weights

The set of samples demonstrates how to deploy above models to run inference with GStreamer command line tool gst-launch-1.0 and Intel® DL StreamerDL Streamer framework components.

Samples

The 'geti_sample.sh' script sample builds GStreamer pipeline composed of the following elements:

• filesrc or urisourcebin or v4l2src for input from file/URL/web-camera
• decodebin3 for video decoding
• gvadetect uses for full-frame object detection and marking objects with labels
• gvaclassify uses for full-frame object classficiation
• gvawatermark for points and theirs connections visualization
• autovideosink for rendering output video into screen
• vah264enc or vah264lpenc and filesink for encoding video stream and storing in a local file

NOTE: sync=false property in autovideosink element disables real-time synchronization so pipeline runs as fast as possible

Before running geti_deployment samples

    export MODELS_PATH="$HOME"/models
    cd /opt/intel/dlstreamer/samples/gstreamer/gst_launch/geti_deployment/

Example deployment of Geti™ anomaly detection UFlow model using CPU device, saving results into a file on local disk.

NOTE: User must provide a relevant input video file.

./geti_sample.sh anomaly-detection uflow/FP32/model.xml CPU opencv INPUT_VIDEO_FILE json

Example deployment of Geti™ bounding-box detection model using GPU device, saving results into a file on local disk.

./geti_sample.sh detection GPU

Example deployment of Geti™ single-label classification model using NPU device, saving results into a file on local disk.

./geti_sample.sh classification_single NPU

Example deployment of Geti™ multi-label classification model using CPU device, saving results into a file on local disk.

./geti_sample.sh classification_multi CPU

See also

• Samples overview