In this folder you can find the data used to train our affordances Bayesian Networks.
For more detail on affordance datasets and on the data acquisition process with the iCub, see also:
- [affExploreManager] (https://github.com/robotology/poeticon/tree/master/src/affExploreManager)
- [3d_pull_push_data] (https://github.com/atabakd/3d_pull_push_data)
- [A Dataset on Visual Affordances of Objects and Tools] by A. Dehban (http://vislab.isr.tecnico.ulisboa.pt/datasets/)
The files pcaTrain_2n2c_2v.m and pcaTrain_2n2c_4v.m are the most important ones, where
- n: number of PCA;
- c: number of components for each PCA;
- v: number of discretization values/bins of each PCA component.
pcaTrain_2n2c_2v.m loads data from Data+combinator folder (change the path in line 15 if you want to use a different data).
The script outputs the trained network as a .mat file in the Networks+listener folder (you may change the filename in line 117).
This folder contains the trained networks and the listener MATLAB YARP script.
The listener (prediction.m) loads one of the networks according to the bn variable (the default value being pca_2016_2n_2c_2v) and waits for queries.
See prediction.m for more detailed information.
Old networks and scripts that were used in the 3rd year demo (March 2015) are kept here for historical reference.
Released under the terms of the GPL v2.0 or later. See LICENSE for details.