Deep Learning with PyTorch

Transposed convolutions

Transposed convolutions are also called fractionally strided convolutions. They work in the opposite way to how convolution works. Intuitively, they try to calculate how the input vector can be mapped to higher dimensions. Let's look at the following figure to understand it better:

This diagram is referenced from Theano (another popular deep learning framework) documentation (http://deeplearning.net/software/theano/tutorial/conv_arithmetic.html). If you want to explore more about how strided convolutions work, I strongly recommend you read this article from the Theano documentation. What is important for us is, that ...