Growing Interpretable Part Graphs on ConvNets via Multi-Shot Learning

This paper proposes a learning strategy that extracts objectpart concepts from a pre-trained convolutional neural network (CNN), in an attempt to 1) explore explicit semantics hidden in CNN units and 2) gradually grow a semantically interpretable graphical model on the pre-trained CNN for hierarchical object understanding. Given part annotations on very few (e. g. 3–12) objects, our method mines certain latent patterns from the pre-trained CNN and associates them with different semantic parts. We use a four-layer And-Or graph to organize the mined latent patterns, so as to clarify their internal semantic hierarchy. Our method is guided by a small number of part annotations, and it achieves superior performance (about 13%–107% improvement) in part center prediction on the PASCAL VOC and ImageNet datasets1.

Authors

• Quanshi Zhang Shanghai Jiao Tong University
• Ruiming Cao
• Ying Nian Wu University of California, Los Angeles
• Song-Chun Zhu State Key Laboratory of General Artificial Intelligence, BIGAI

Keywords

No keywords are indexed for this paper.