Artificial Intelligence Stack Exchange is a question and answer site for people interested in conceptual questions about life and challenges in a world where "cognitive" functions can be mimicked in purely digital environment. It only takes a minute to sign up.
Sign up to join this community
I see some papers use the term "natural image domain". I googled that but didn't find any explanation of it.
I guess I understand the normal meaning of "natural image", such as the image people take by phones. The images in ImageNet database are all natural images.
Is "natural image domain" a subfield of computer vision?
An image of a natural domain refers to any kind of image that has some variance in its structure, so it does not always present the same peculiarities.
For example, in a digits classification task, you can train your network on the MNIST dataset, that is a simple black & white set of images that are always well defined: the digit is in darker pixels, and the background is always white, they just have different shapes.
For the same task but in the natural image domain, the digits may vary in color, noise, and even position inside of a single image. As you may imagine this does not apply to digits only, but it includes any image that was not specifically created or modified to be fed into a neural network.
So, to put it simply, a neural network trained for natural images is able to perform classification in any real-life situation or scenario, without the need to edit the image to adapt it to what the network was trained for (for example, changing the color or cropping so the object you want to recognize is exactly in the center).
Here a similar question that may fill some doubts you may have.
