If MobileNet is like that, we can know where to cut our network. Just re-use the convolutional layers.
%% Cell type:markdown id: tags:
This above printed (I think) the structure of the network.
What's particularly interesting is the last layer of the network, called classifier.
Sequential() seems to be just a way to "link" layers together (see [Sequential doc](https://pytorch.org/docs/stable/nn.html?highlight=sequential#torch.nn.Sequential)). The "classifier" is composed of two layers:
1. Dropout layer: which during training zeroes randomly some elements of the input tensor (This is good for the
2. Linear layer: Linear transformation of incoming data
This seems a bit stiff to use fo classifying our pictures (I was hoping for a fully connected layer) but ok.
%% Cell type:code id: tags:
``` python
# Download an example image from the pytorch website