In this paper, we study the problem of semantic part segmentation foranimals. This is more challenging than standard object detection, objectsegmentation and pose estimation tasks because semantic parts of animals oftenhave similar appearance and highly varying shapes. To tackle these challenges,we build a mixture of compositional models to represent the object boundary andthe boundaries of semantic parts. And we incorporate edge, appearance, andsemantic part cues into the compositional model. Given part-level segmentationannotation, we develop a novel algorithm to learn a mixture of compositionalmodels under various poses and viewpoints for certain animal classes.Furthermore, a linear complexity algorithm is offered for efficient inferenceof the compositional model using dynamic programming. We evaluate our methodfor horse and cow using a newly annotated dataset on Pascal VOC 2010 which haspixelwise part labels. Experimental results demonstrate the effectiveness ofour method.
展开▼