################################################################### # # # LSML Toolbox Version 1.10 # # (Matlab code for Locally Smooth Manifold Learning) # # http://www.vision.ucsd.edu/~pdollar # # # # Piotr Dollar (pdollar-at-caltech.edu) # # Vincent Rabaud (vrabaud-at-cs.ucsd.edu) # # # ################################################################### 1. Introduction. This Matlab software library implements algorithms described in: @inproceedings{ ICML07, author = "P. Doll\'ar and V. Rabaud and S. Belongie", title = "Non-Isometric Manifold Learning: Analysis and an Algorithm", booktitle = "ICML", month = "June", year = "2007" } @inproceedings{ NIPS06, author = "P. Doll\'ar and V. Rabaud and S. Belongie", title = "Learning to Traverse Image Manifolds", booktitle = "NIPS", month = "Dec.", year = "2006" } If you use this software for research purposes, you should cite the aforementioned paper in any resulting publication. ################################################################### 2. License. This toolbox is published under the LGPL license. Please read lgpl.txt and gpl.txt for more info. ################################################################### 3. Installation. This toolbox requires "Piotr's Image & Video Toolbox for Matlab", version 2.20 or higher, available from: http://vision.ucsd.edu/~pdollar/toolbox/doc/index.html Follow the download/install instruction for the above first. It is crucial that the C++ code for Dijkstra's algorithm compiles or the included binaries work properly (run the example in dijkstra.m to check). Finally, unzip the LSML toolbox and add to Matlab's path. Tested under Matlab 2007b/2008a/2008b on Windows/Linux. ################################################################### 4. Getting Started. A number of demos have been prepared to familiarize users with the LSML code. To run them, from Matlab type: >> lsmlDemo(1) % Generate several data sets and display them. >> lsmlDemo(2) % Compute nearest neighbors and display them. >> lsmlDemo(3) % Learn tangents on a manifold. >> lsmlDemo(4) % Compute geodesic distances on a manifold. >> lsmlDemo(5) % Compute low-dim embedding of a manifold. >> lsmlDemo(6) % Denoise points on a manifold. Also see the Contents.m file for function summaries and the help of individual m-files for more information. ################################################################### 5. Contents. For a list of toolbox contenst see Contents.m. ################################################################### 6. History / ToDo. Version 1.10 (06-Mar-09) -significant speedups (up to 10x) for learning -note: requires update to dependent toolboxes Version 1.01 (17-Nov-08) -removed dependencies on obscure Matlab toolboxes (thanks to Junbin Gao for pointing these out) -tested with Matlab 2008b on Windows Version 1.00 (08-July-08) -initial version ToDo: -more efficient embedding code (lsmlEmbed) ###################################################################