From Wang et al (2015) "Visual Tracking with Fully Convolutional Networks":

A top layer encodes more semantic features and serves as a category detector, while a lower layer carries more discriminative information and can better separate the target from distracters with similar features.

I don't quite understand this statement: does this mean that the algorithm first identifies the semantic group to which the object belongs and then tries to distinguish among them? Also, are there any other publications where the difference in the roles of layers been demonstrated?


1 Answer 1


This is a basic property of convolutional networks. First layers identify simple features and as you go deeper in the CNN each layer uses the features of the previous layer to build more complex ones.

I would say the state-of-the-art article about this is Visualizing and understanding Convolutional Networks by Matthew D. Zeiler and Rob Fergus. You can download it here


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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