The performance in object detection of features extracted from the CFN is the highest (51.8%) among unsupervisedly trained features, and very close to that of supervisedly trained features (56.5%).