This project is a Matlab implementation of Trottier, L. et al. (2016) Dictionary Learning for Robotic Grasp Recognition and Detection.
The Cornell dataset and the sub-sampled Washington RGBD dataset used in the experiments of Trottier, L. et al. (2016) is already available in folder data.
To install:
- Install the toolboxes by running file setup.m
- Add current folder and all sub-folders to path with
addpath(genpath('.'));
File recognitionMain.m, available in folder recognition, is a detailed example
on how to perform dictionary learning, sparse coding and SVM training with cross-validation.
To see all available options for additional tweaking, see file loadOverallParameters.m
in folder opts.
Reproducing Trottier, L. et al. (2016)'s Experiments
Folder tasks/recognition contains all recognition experiments used to produce
the results presented in Trottier, L. et al. (2016).
For instance, running taskRpOmp.m perform 5-5 nested cross-validation using
normalized random patches as dictionary and Orthogonal Matching Pursuit as features.
By default, the script is verbose. Consider setting opts.util.verbose = false
to turn off verbose mode.