jizhihang / aogdetector

Source code for object detection with And-Or Graphs

AOGDetector

Implementation of the papers

    Xi Song, Tianfu Wu, Yunde Jia and Song-Chun Zhu, "Discriminatively Trained And-Or Tree Model for Object Detection",  In CVPR2013.        
    Bo Li, Wenze Hu, Tianfu Wu and Song-Chun Zhu, "Modeling Occlusion by Discriminative AND-OR Structures", In ICCV2013        
    Bo Li, Tianfu Wu and Song-Chun Zhu, "Integrating Context and Occlusion for Car Detection by Hierarchical And-Or Model", In ECCV2014
    Tianfu Wu, Bo Li and Song-Chun Zhu, "Learning And-Or Models to Represent Context and Occlusion for Car Detection and Viewpoint Estimation",  arXiv 1501.07359, TPAMI(under major revision), 2015

Copyright (c) 2014 Center for Vision, Cognition, Learning and Autonomy at UCLA http://vcla.stat.ucla.edu/,

    Written by Matt Tianfu Wu (primary contact) <[email protected]>,  
    Xi Song                          <[email protected]> (object detection module),                        
    Bo Li                            <[email protected]>  (occlusion module),                      

Code Release Log

 10-14-2014: Release detection code  (full code will be released in future)                    

MODULE 1: object detection

VOC Configuration

All the configuration can be specified in voc_config.xml in "<PATH_TO_RGM_Source_Code>/config" folder.

VOC Testing

Use Terminal and cd to the <PATH_TO_Build>/bin where the executable "Entry" is built,

Run: Entry vocEval PATH_To_voc_config_xml,

e.g., Entry vocEval /home/tfwu/Code/RGM/config/voc_config.xml

DetectionInASingleImage

Use Terminal and cd to the PATH_TO_Build/bin where the executable "Entry" is built,

Run: Entry detect PATH_To_voc_config_xml,

e.g., Entry detect /home/tfwu/Code/RGM/config/voc_config.xml

BatchDetectionInDirectory

Use Terminal and cd to the PATH_TO_Build/bin where the executable "Entry" is built,

Run: Entry detectBatch PATH_To_voc_config_xml,

e.g., Entry detectBatch /home/tfwu/Code/RGM/config/voc_config.xml

ConvertVOCRel5Model

Use Terminal and cd to the <PATH_TO_Build>/bin where the executable "Entry" is built,

Run: Entry cvtVOCRel5Model PATH_To_VOCRel5_Model_Mat_File,

e.g., Entry cvtVOCRel5Model /home/tfwu/Code/RGM/externalModels/VOC2007/bicycle_final.mat,

The converted model will be saved (e.g., /home/tfwu/Code/RGM/data/bicycle_final.mat.bin) which can be used to test detection.

Note: Before using this to convert a matlab model trained using VOC-Release 5 by Dr. Ross Girshick, please use PATH_TO_RGM_Source_Code/matlab/convertVOCR5.m to add a "DFS" field to the model structure.