Adding custom udfs in VideoIngestion/VideoAnalytics¶

For python udfs: Copy udf code to [WORKDIR]/IEdgeInsights/common/video/udfs/python directory

Refer the below example to copy the python safety gear udf (For this example python safety_gear udf can be copied from [WORKDIR]/IEdgeInsights/CustomUdfs/PySafetyGearAnalytics):

$ cp -r /safety_gear [WORKDIR]/IEdgeInsights/common/video/udfs/python

For native udfs: Copy udf code to [WORKDIR]/IEdgeInsights/common/video/udfs/native directory

Refer the below example to copy the native safety gear udf (For this example native safety_gear_demo udf can be copied from [WORKDIR]/IEdgeInsights/CustomUdfs/NativeSafetyGearAnalytics):

$ cp -r /safety_gear [WORKDIR]/IEdgeInsights/common/video/udfs/native
Make sure the appropriate ingestion config for the udf is set. In case one needs to use a video file then copy the video file to [WORKDIR]/IEdgeInsights/VideoIngestion/test_videos and make the config changes.

Refer the below example to copy Safety_Full_Hat_and_Vest.avi video to [WORKDIR]/IEdgeInsights/VideoIngestion/test_videos

$ cp -r Safety_Full_Hat_and_Vest.avi [WORKDIR]/IEdgeInsights/VideoIngestion/test_videos

In [WORKDIR]/IEdgeInsights/VideoIngestion/config.json make the below changes to use Safety_Full_Hat_and_Vest.avi video file for safety_gear udf
```
----snip----
"ingestor": {
         "type": "opencv",
         "pipeline": "./test_videos/Safety_Full_Hat_and_Vest.avi",
         "loop_video": true,
         "queue_size": 10,
         "poll_interval": 0.2
     },
```

Make the required config changes to add the udfs.

For python udfs:

Refer the below example to make the required config changes for python safety gear udf:

In [WORKDIR]/IEdgeInsights/VideoIngestion/config.json or [WORKDIR]/IEdgeInsights/VideoAnalytics/config.json make the below changes to use safety_gear udf

----snip----
  "udfs": [
      {
         "device": "CPU",
         "model_bin": "common/video/udfs/python/safety_gear/ref/frozen_inference_graph.bin",
         "model_xml": "common/video/udfs/python//safety_gear/ref/frozen_inference_graph.xml",
         "name": "safety_gear.safety_classifier",
         "type": "python"

     }
  ]

For native udfs:

Refer the below example to make the required config changes for native safety gear udf:

In [WORKDIR]/IEdgeInsights/VideoIngestion/config.json or [WORKDIR]/IEdgeInsights/VideoAnalytics/config.json make the below changes to use safety_gear udf

----snip----
  "udfs": [
      {
         "device": "CPU",
         "model_bin": "common/video/udfs/native/safety_gear_demo/ref/frozen_inference_graph.bin",
         "model_xml": "common/video/udfs/native/safety_gear_demo/ref/frozen_inference_graph.xml",
         "name": "safety_gear_demo",
         "type": "native"

     }
  ]

Note: The IR model files must be copied along with udfs

For native udfs one needs to build it by modifying the [WORKDIR]/IEdgeInsights/common/video/udfs/native/CMakeLists.txt file to add the subdirectory of the udf for build.

Refer the below example to modify the [WORKDIR]/IEdgeInsights/common/video/udfs/native/CMakeLists.txt to build the native safety_gear_demo
```
----snip----
add_subdirectory(safety_gear_demo/)
----snip----
```
Note: For custom udfs a CMakeLists.txt file needs to be created and must reside in the udf directory which is added as a subdirectory to [WORKDIR]/IEdgeInsights/common/video/udfs/native/CMakeLists.txt file.
Run builder and provision by refering [WORKDIR]/IEdgeInsights/README.md
Build and run VideoIngestion & VideoAnalytics services.