concept image tensor in category deep learning

This is an excerpt from Manning's book Deep Learning with JavaScript: Neural networks in TensorFlow.js. Login to get full access to this book.

If the output is regarded as an image tensor (a totally valid way of looking at this!), then filters can be understood as the number of channels in the output. Unlike the input image, the channels in the output tensor don’t actually have to do with colors. Instead, they represent different visual features of the input image, learned from the training data. For example, some filters may be sensitive to straight-line boundaries between bright and dark regions at a certain orientation, while others may be sensitive to corners formed by a brown color, and so forth. More on that later.

tf.browser.fromPixels: Getting image tensors from HTML img, canva- as, or video elements

The second way to get image tensors in the browser is to use the TensorFlow.js function tf.browser.fromPixels() on HTML elements that contain image data—this includes img, canvas, and video elements.