Fig. 5: Structure and performance of UPENN and Contrast Net.

a Schematic diagram of UPENN and cross-layer transmission, marked with pre-learning and re-learning methods. b, c Gradient distribution calculated by selecting specific layers of stage 1 and stage 4 in UPENN and Contrast Net. d Feature maps extracted from the same photo in stage 1 and stage 2 of two networks. e The feature maps extracted from the same photo in stage 3 of two networks, the content of Contrast Net has lost most of its features. f Characteristic Diagram in stage 4 of Two Networks. g The recognition accuracy of UPENN and Contrast Net on the unlearned content after pre-learning. h The recognition accuracy of UPENN and Contrast Net after re-learning.