Can YOLOv3 architecture be clearly separated into feature extractor and classifier parts?

Question

I am new to machine learning and am confused about whether its architecture has clearly defined boundaries which demarcate the feature extraction and classification part. Or is it that it classifies the features extracted at different scales throughtout the network ?

Answer 1

The feature extraction part is described in section 2.4 of the original paper. The proposed darknet-53 is an improvement over the previous darknet-19 which didn't include residual layers.

After the darknet backbone, three detection blocks are responsible for predicting bounding boxes and probability scores for each object class. notice that YOLO is an end-to-end architecture, so bounding boxes and objects are predicted at the same time.