build.sh not working out of the box.

Will be investigating this further, but as of now, the instructions given in the README.md and samples/README.md do not seem to work out-of-the-box:

README.md

Building

To get started, build the service as a standalone component execute the following command

./docker/build.sh

samples/README.md

1. Build the service container with sample pipelines:

   ~/video-analytics-serving$ ./docker/build.sh
   

When running build.sh the following error is thrown:

./docker/build.sh: line 165: ${MODELS^^}: bad substitution

I’m assuming necessary flags aren’t being set, but in such case, I believe that the README’s must be updated. Will update this thread further into the night after some code-reading.

[nnshah1]

You should be able to increase the log_level to DEBUG to get more info on the exact error. Most probably in this case there is a typo in the pipeline.

We are working on improving the documentation, but we can help you through. The basics are to add a directory under models with the structure: model_name/version/FP32/*.bin, *.xml. Once the model is there you can refer to it in a pipeline by: models[model_name][version][network] and VA Serving will fill in the full path at runtime.

The challenge with this specific model is that it’s post-processing is not currently handled directly by DLStreamer and you will have to write some custom processing code to convert the tensor output into a bounding box. The simplest approach would be to use the gvapython element and implement the post processing in python. (Section 3 here)

https://github.com/opencv/gst-video-analytics/wiki/Custom-Processing

The post processing for this model can be found here:

https://github.com/opencv/open_model_zoo/blob/master/demos/text_detection_demo/src/text_detection.cpp

Many models (ssd, tiny lolo, mobilenet-v2 classification) work out of the box - but unfortunately not this one (yet). If you’d like to see this feature please add a request on the DLStreamer repo.

https://github.com/opencv/gst-video-analytics

If you decide to go the custom processing route - let us know if you run into any issues

[nnshah1]

We will double check the shell requirements and try to find a way to make it portable - in the mean time let us know if those commands work - thanks!