What is a feature in computer vision? I am new to computer vision. I know that often Convolutional Layers are used and that they apply a filter(/kernel) w over the input x of the layer:
z = w*x + b (b: bias)

*

*Did I understand correctly that the feature is the output of the layer after applying the filter to the input x?

*If that is true, what is the difference between a feature and a feature map?

Thank you and best regards
 A: Feature in computer vision means different things in different contexts. In general, there is no difference in the definition of a feature between computer vision and other machine learning problems. The feature is "individual measurable property or characteristic of a phenomenon being observed".
In tabular data, this is a column of recorded or computed values. In natural language processing, a feature may be one-hot code for a letter or a word, TF-IDF, embedding vector computed using GloVe, or BERT, etc. Same as with passing raw text through GloVe in NLP, using things like convolutional layers in computer vision transforms the raw data, and we can think of this transformed data as features processed by the later layers of the network. Features usually extract meaning or information from the data.
Neural networks gained popularity because they do automated feature engineering. Before neural networks you would need something like an algorithm, that would extract information about edges on the picture so that some machine learning algorithm would use the knowledge about edges, to learn something from the data. Neural networks figure it out by themselves, to at least we hope they will do it for us.
For the definition of the feature map, check the What is the definition of a "feature map" (aka "activation map") in a convolutional neural network? thread. TL;DR you are correct, feature map is the output of the previous layer of the network, a set of features learned by it.
