Convolutional Network terminologies

ConvNet Terminologies.JPG

  1. Patch / kernel : size of the small image tile
  2. Stride : The number of pixels we will be shifting each time we move the filter.
    Stride of 1 makes the output the same size as input. The stride of two makes the output roughly half the size of the inputConvNet Terminologies -stride.JPG
  3. Padding:
    1. Valid Padding : If we do not go off the edge
    2. Same Padding : If the output map size is exactly the same as input. For the extra rows or columns they are padded with the zeroes in c this casevalid vs same padding.JPG
  4. Depth = semantic representation of the complexity i.e more depth = more complex learning. Increase the depth layer after layer in case of ConvNets

Finding the stride, depth  with valid and same padding

stride -depth and padding.JPG

Advertisements