{%hackmd SybccZ6XD %}
###### tags: `paper`

# Spatial Transformer Networks

What is that?
> The spatial transformer module is a **dynamic mechanism** that can actively spatially transform an image

The architecture
> 
> U: input feature map
> V: output feature

Localisation net
> Predict the parameter of $T_\theta(G)$
> 

The example of $T_\theta(G)$
> 

Why it need the sampler
> The output and input is the same size, so we only need to sample certain part.
> 

## Appendix

We define an MNIST addition task, where the network must output the sum of the two digits given in the input.
> 

The classification task
> Find the feature part and predict
>