Please enable JavaScript.
Coggle requires JavaScript to display documents.
CNN - Coggle Diagram
CNN
-
-
Filters, Kernel and Parameters
-
A filter is composed by a number of kernels equals to the number of channels in input. so we have a kernel for channel. If filters is greater then 1 we'll have multichannel output 
-
Stride
How much the kernel moves (horizontally and vertically) when it pass through the image. s=1 means pixel by pixel, s=2 it jumps a pixel so moves of 2 pixel at times and so on
-
Residual Network
pass to internal levels also the informations from less internal levels to avoid vanishing gradient and helping the error to converge
-
U-Net
Architecture used for image segmentation. It is characterized by a contraction phase, followed by an expansion phase
Autoencoder
It builds similar data (decoder) very similar to original data and cleans the data from noise. It learns in an unsupervised way. It ois formed by 2 distinct networks: Encoder and Decoder
-
-
-
-
-
Dropout
It reduces overfitting eliminating randomly neurons from NN. Nackpropagation is done on the diminished network.
-
-
-
-
-
-
-
-
-
-