Litcius/Paper detail

Extraction of an Explanatory Graph to Interpret a CNN

Quanshi Zhang, Xin Wang, Ruiming Cao, Ying Wu, Feng Shi, Song‐Chun Zhu

2020IEEE Transactions on Pattern Analysis and Machine Intelligence37 citationsDOIOpen Access PDF

Abstract

in a conv-layer usually represents a mixture of object parts. We develop a simple yet effective method to learn an explanatory graph, which automatically disentangles object parts from each filter without any part annotations. Specifically, given the feature map of a filter, we mine neural activations from the feature map, which correspond to different object parts. The explanatory graph is constructed to organize each mined part as a graph node. Each edge connects two nodes, whose corresponding object parts usually co-activate and keep a stable spatial relationship. Experiments show that each graph node consistently represented the same object part through different images, which boosted the transferability of CNN features. The explanatory graph transferred features of object parts to the task of part localization, and our method significantly outperformed other approaches.

Topics & Concepts

Computer scienceGraphArtificial intelligencePattern recognition (psychology)Convolutional neural networkObject (grammar)Feature extractionGraph theoryTheoretical computer scienceMathematicsCombinatoricsAdvanced Neural Network ApplicationsAdversarial Robustness in Machine LearningAnomaly Detection Techniques and Applications
Extraction of an Explanatory Graph to Interpret a CNN | Litcius