Deep Learning Essentials

Convolution layer

One of the core building blocks of CNNs, the convolutional layer is responsible for applying a specific convolution filter to an input image. This filter is applied on each sub-region of the image, which is further defined by the local connectivity parameter of the layer. Each filter application produces a scalar value for a specific pixel location, which when combined across all pixel locations is often referred to as a feature map. For example, if you use eight filters to convolve a 32 x 32 image at every single pixel location, you will produce 12 output feature maps each of the size 32 x 32. In this case, each of the feature maps will be computed corresponding to a particular convolution filter. The Example of a convolution ...