Predicting an image class using MobileNet V2

We have previously discussed how running Inception V3 gives us outstanding results on the ImageNet dataset, but sometimes the inference is considered to be slow. Meet MobiletNet V2, a neural networks architecture developed to deliver excellent results within a short period of time.

Setting up the environment

Unfortunately, MobileNet V2 is not present in the MXNet Model Zoo. We will use two different GitHub repositories to get the symbol, params, and synset files.

Firstly, please navigate to the KeyKy Mobile Net repository (https://github.com/KeyKy/mobilenet-mxnet) and download the following two files:

• mobilenet_v2-0000.params
• mobilenet_v2-symbol.json

Next, navigate to the MobileNet Caffe implementation (https://github.com/shicai/MobileNet-Caffe) and download synset.txt.

Place all of the downloaded files into the weights/mobilenet-v2 folder.

Loading the network

You previously loaded Inception V3. The process for this is the same. We will use the mx.load_checkpoint...