concept color channel in category deep learning
This is an excerpt from Manning's book Deep Learning with PyTorch.
The dimensions (or axes) of our tensors usually index something like pixel locations or color channels. This means when we want to index into a tensor, we need to remember the ordering of the dimensions and write our indexing accordingly. As data is transformed through multiple tensors, keeping track of which dimension contains what data can be error-prone.
We mentioned colors earlier. There are several ways to encode colors into numbers.1 The most common is RGB, where a color is defined by three numbers representing the intensity of red, green, and blue. We can think of a color channel as a grayscale intensity map of only the color in question, similar to what you’d see if you looked at the scene in question using a pair of pure red sunglasses. Figure 4.1 shows a rainbow, where each of the RGB channels captures a certain portion of the spectrum (the figure is simplified, in that it elides things like the orange and yellow bands being represented as a combination of red and green).