Hierarchical Matching Pursuit

Multipath Hierarchical Matching Pursuit achieves state-of-the-art results on many types of recognition tasks. Linear SVM is sufficient for good accuracy. Please check our demo code in the package. If you use our package, please cite the following papers.

Source code

Publications

Learning Algorithms for Recognition

  • Liefeng Bo, Xiaofeng Ren and Dieter Fox, Multipath Sparse Coding Using Hierarchical Matching Pursuit, In IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), June 2013. [PDF] [BIB]

  • Liefeng Bo, Xiaofeng Ren and Dieter Fox, Unsupervised Feature Learning for RGB-D Based Object Recognition, International Symposium on Experimental Robotics (ISER), June 2012. [PDF] [BIB]

  • Liefeng Bo, Xiaofeng Ren and Dieter Fox, Hierarchical Matching Pursuit for Image Classification: Architecture and Fast Algorithms, Advances in Neural Information Processing Systems (NIPS), December 2011. [PDF] [BIB]

Attribute Based Object Identification, RGB-D Surface Model Compression and Contour Detection

  • Yuyin Sun, Liefeng Bo and Dieter Fox, Attribute Based Object Identification, In IEEE International Conference on Robotics and Automation (ICRA), May, 2013. [PDF] [BIB]

  • Michael Ruhnke, Liefeng Bo and Dieter Fox and Wolfram Burgard, Compact RGBD Surface Models Based on Sparse Coding, In the AAAI Conference on Artificial Intelligence (AAAI), July 2013. [PDF] [BIB]

  • Xiaofeng Ren and Liefeng Bo, Discriminatively Trained Sparse Code Gradients for Contour Detection, Advances in Neural Information Processing Systems (NIPS), December, 2012. [PDF] [BIB] [Code]

The core of building recognition systems is to extract expressive features from high-dimensional structured data, such as images, depth maps, 3D point clouds, videos and audios. Hierarchical Matching Pursuit (HMP) aims to discover such features from raw sensor data. As a multilayer sparse coding network, HMP builds feature hierarchies layer by layer with an increasing receptive field size to capture abstract features. HMP uses sparse coding to learn codebooks at each layer in an unsupervised way and builds hierarchial feature representations from the learned codebooks using orthogonal matching pursuit, spatial pooling and contrast normalization. HMP achieves state-of-the-art results on many types of recognition tasks. Linear classifiers are sufficient for good accuracy.

This work was funded in part by the Intel Science and Technology Center for Pervasive Computing and by ONR MURI grant N00014-07-1-0749.