Efficient 3D Scene Labeling Using Fields of Trees

摘要: We address the problem of 3D scene labeling in a structured learning framework. Unlike previous work which uses Support Vector Machines, we employ recently described Decision Tree Field and Regression frameworks, learn unary binary terms Conditional Random from training data. show this has significant advantages inference speed, while maintaining similar accuracy. also demonstrate empirically importance for overall accuracy features that make use prior knowledge about coarse layout such as location ground plane. how can be estimated by our framework automatically, information used to bootstrap improved detailed labeling.