The Aggregated Partition for Discovering Objects

INTRODUCTION:

The Aggregated Partition (AP) and the Selective AP are two models for the problem of unsupervised object discovery in image collections.

This is a repository with an implementation of the models described in our CVPR2016 Workshop paper. We provide here the codes and data needed to reproduce all the experiments detailed in the paper.

License

AP is released under the MIT License (refer to the LICENSE file for details).

CITING

If you make use of this data and software, please cite the following reference in any publications:

@inproceedings{LopezSastre2016,
    Title                    = {Unsupervised Robust Feature-based Partition Ensembling to Discover Categories},
    Author                   = {Lopez-Sastre, R.~J.},
    Booktitle                = {CVPR Workshops},
    Year                     = {2016}
}

REQUIREMENTS:

The AP code is developed and tested under Ubuntu 14.04. Matlab is required.

COMPILING:

Before running any experiment, you should follow the following instructions:

Mex files

Open Matlab and compile the following two mex files:

    cwd=cd; %local path where you have installed AP release
    cd code/generic_functions/
    mex chi_squared_c.c
    cd(cwd);
    cd code/code_weights/
    mex compute_distances_c.c
    cd(cwd);

Clustering Aggregation THIRD-PARTY SOFTWARE

    cd code/aggregation/ca_code
    make all   

---

DISCOVERING CATEGORIES:

We provide in the folder "experiments" the corresponding Matlab scripts to use the AP models in the three datasets described in our paper.

For example, to discover categories in the Caltech-256, simply, open Matlab and go to the corresponding folder and run the provided script.

• For the AP model:

   cd experiments/Caltech-256/SpectralClustering/AP
   experiment_ap   

• For the Selective AP:

   cd experiments/Caltech-256/SpectralClustering/SAP
   experiment_selective_ap 

---