1
$\begingroup$

I'm learning how U-NET network works to do semantic segmentation.

I think I have understood everything but features. What are those image features?

I read that convolutional layers extract features from the images using their filters, but what are they? Are they corners? edges? colours?

I have read this article, "Finding Features", but I think I need more information about them.

$\endgroup$
3
$\begingroup$

CNNs like U-Net extract lower level features like edges on lower layers (i.e. the first convolutional layers) and higher level features on higher layers (i.e. convolutional layers closer to the final linear layers). This principle is losely inspired by how visual perception is implemented in the Visual Cortex among humans (and other animals).

In a CNN the feature maps could for example look like this:

CNN feature maps

As you can see the lower level feature maps detect simple structures like edges while higher level feature maps recognize more complex structures like eyes or faces.

Colors, however, are processed differently by CNNs than spatial features since color images are usually fed to CNNs using three input channels (one for each color in RGB format). So colors are not detected in the same way as spatial features but instead the first convolutional layer receives a 3-dimensional input image with one dimension for each color-component (one for red, one for green, one for blue).

The article What exactly does CNN see? and the paper Visualizing and Understanding Convolutional Networks (which are also the sources of above images) provide a more detailed explanation.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.