annInferenceApp - Client Application

This Sample Inference Application supports:

• connect to inference_server_app

  • queries number of GPUs on the server

  • queries pre-loaded neural network models on the server

  • upload pre-trained CAFFE models (optional)

  • specify input width and height dimensions

  • browse and pick deploy.prototxt and weights.caffemodel files

  • specify input preprocesor values for normalization (if needed)

  • specify input channel order (RGB or BGR)

  • optionally, you can publish your neural network for use by others

    • select “PublishAs”

    • give a name to your model

    • type password (default passowrd of inference_server_app is “radeon”)

• pick a CNN model, if upload option is not used

• run inference on images from a folder

  • browse and pick labels (i.e., synset words for each output label)

  • browse and pick input images to be classified

  • optionally, pick list of input images from a .txt or .csv file

  • optionally, specify max number of images to be processed from folder or image list

If you want to run a simple test, use annInferenceApp.py (a python script) to simply pick a network model to classify local images.

% python annInferenceApp.py [-v] -host:<hostname> -port:<port> -model:<modelName>
                            [-synset:<synset.txt>] [-output:<output.csv>]
                            <folder>|<file(s)>

• Using QMake

  If you want to avoid installing QtCreator application (Qt creator app is only needed for debugging and designing purposes), you can also build the app using QMake

    sudo apt-get install qt5-qmake
    cd MIVisionX/apps/cloud_inference/client_app/
    qmake -o Makefile annInferenceApp.pro
  

  The above qmake command will generate a Makefile for the application, which can be used to build the application. When your application is successfully built, run the app as you are running ordinary c application.

    make
    ./annInferenceApp