Integrating Context and Occlusion for Car Detection by Hierarchical And-Or Model

Bo Li1,2, Tianfu Wu2 and Song-Chun Zhu2

1Beijing Lab of Intelligent Information Technology, Beijing Institute of Technology

2Department of Statistics, University of California, Los Angeles


We present a reconfigurable hierarchical And-Or model to integrate context and occlusion for car detection in the wild. The model structure is learned by mining context and viewpoint-occlusion patterns at three levels: a) N-car layouts, b) single car and c) car parts. Our model is a directed acyclic graph (DAG) where dynamic programming (DP) algorithm can be used in inference. The model parameters are learned by weak-label Structural SVM. Experimental results show that our model is effective in modelling context and occlusion information in complex situations, and obtains better performance over state-of-the-art car detection methods.

Hierarchical And-Or Model




  • Parking-Lot Dataset: This file contains training and testing data (with bounding box annotations) described in our ECCV 2014 paper.
  • Detection Code: This file contains the detection code and pre-trained models described in our ECCV 2014 paper.
  • Poster: This is our ECCV 2014 poster.