Hello, I am using a convolutional decoder like network to recreate 500x500 point images (continuous data between 0,1) from 3 points of data. My network is not too complex and looks like the image shown below, yet with a little under 3000 training samples and a batch size of 300 it takes over six hours for a single epoch. This is rediculous! Any idea why?
The Torch layers and activations i’m using are ConvTransposed2d and 3d, PReLU, Linear, and Tanh.