How to retrain a detection model and quantize it for Intel Movidius?

I want to retrain an existing object detection model with a new image dataset and quantize it for Intel Movidius. Is there any working procedure to do this?

I have successfully retrained the model but failing to quantize it. I have followed the following tutorial Retrain SSD MobileNet

The Movidius devices only support FP16 models, and to convert a caffe version of SSD Mobilenet then you supply the "--data_type FP16"to the model optimizer (mo.py) The openvino model zoo has a mobilenet-ssd model,also using caffe, and the associated yaml file has the following parameters model_optimizer_args:

• --input_shape=[1,3,300,300]
• --input=data
• --mean_values=data[127.5,127.5,127.5]
• --scale_values=data[127.5]
• --output=detection_out
• --input_model=$dl_dir/mobilenet-ssd.caffemodel
• --input_proto=$dl_dir/mobilenet-ssd.prototxt.

Note that your input shape and the mean and scale values will likely be different so change those to match your retrained model.

There's also a demo file shipped with openvino that can be used with your converted model. See the associated mode.lst file for all the supported architectures. https://github.com/openvinotoolkit/open_model_zoo/tree/master/demos/object_detection_demo/python