1

So I've seen a convolutional neural network that uses the traditional convolutional layers, but then after the down-sampling of the input and some pooling layer, they've added a transposed convolution layer which up-sample the resulting input.

without any context I'm wondering why such thing could be beneficial/useful to add a transposed convolution layer? and if so when should one use this?

I know my question is out of any context but I thought that maybe there is some general reason behind this.

MercyDude
  • 111
  • 5
  • Can you share more context? Upsampling using transposed convolutions is used in many places, for example GANs and autoencoders for generative image modeling and U-Nets for semantic segmentation – Artem Mavrin Aug 24 '20 at 17:39
  • @ArtemMavrin So I've found a CNN for classifying some vectors of brain activity to complex spikes or simple spikes, and this network is using this technique of down-sampling with the convolution layer and then upsampling back again, and this network gets pretty good results, I want to know why. – MercyDude Aug 25 '20 at 20:13

0 Answers0