Mastering Computer Vision with TensorFlow 2.x

Convolution over volume – 3 x 3 filter

In the preceding example, we applied a 3 x 3 convolution over a two-dimensional image (grayscale). In this section, we will learn how a three-dimensional image with three channels (Red, Green, and Blue (RGB)) is transformed by using a 3 x 3 edge filter using a convolution operation. The following diagram shows this transformation graphically:

The preceding diagram shows how a section of a 7 x 7 image's graph can be transformed using a 3 x 3 filter (edge detector) in terms of width reduction and depth increase (from 3 to 32). Each of the 27 (3 x 3 x 3) cells in the kernel (f_i) is multiplied by the corresponding ...