MobileNetV2 demo in pytorch:
  https://pytorch.org/hub/pytorch_vision_mobilenet_v2/

MobileNet V2 paper:  https://arxiv.org/pdf/1801.04381.pdf

Articles with good illustrations:
  https://predictiveprogrammer.com/famous-convolutional-neural-network-architectures-2/
  https://towardsdatascience.com/types-of-convolutions-in-deep-learning-717013397f4d
  https://towardsdatascience.com/visualizing-convolution-neural-networks-using-pytorch-3dfa8443e74e

Inverted residuals:
  https://towardsdatascience.com/mobilenetv2-inverted-residuals-and-linear-bottlenecks-8a4362f4ffd5

MobileNetV2 parameters: t,c,n,s
  t = expansion ratio (nhiddens per input channel) for bottleneck layers
  c = number of input channels
  n = repetition count of blocks (sum of n's = # layers)
  s = stride

================

Files included in this folder:

labels.py holds all 1000 Imagenet class labels.

images/ folder holds the images used by mb_demo1 and mb_demo2.

mb_demo1 loads several images and gives their top 5 classifications.

mb_demo2 loads several images and gives their feature vector
representations.  It doesn't display anything.

mb_demo3 displays all 32 layer 1 kernels, either as blended 3x3 grids
or as three 3x3 opponent color grids.  It then runs through all 32 kernels
and shows the response to a test image.

MobileNet.fsm runs the current camera image through the network and
prints the top 5 classifications.