Convolutional Recurrent Neural Networks
In this post we will be writing convolutional recurrent neural networks in pytorch.
CRNNs are combination of convolutional neural networks and RNNS.
Following is how we can write a CNN in pytorch -
Now we will combine this CNN with LSTM to get a CRNN.
CRNNs gives us an ability to extract spatial as well as temporal features. A simple use-case of CRNNs can be to get numbers from vehicle number plates. CRNNs have also been used to study fMRIs(see here).
CRNNs are really useful when you have to extract high level local features across time.