This paper proposes a learning strategy that extracts object-part 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., 3ā€”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%ā€”107% improvement) in part center prediction on the PASCAL VOC and ImageNet datasets.
In Proceedings of AAAI-17

Recent Posts

I migrated my original site hosted on Google Sites to here on Github Pages. It will hopefully provide better accessibility for visitors from regions where Google has been blocked (i.e. China).

Full Article


Email: caoruiming[at]gmail[dot]com