Detects social groups in sets of person percepts.
> git clone
> mkdir -p fformation/build && cd fformation/build
> cmake .. && make
Allowed options:
-h [ --help ] produce help message
-c [ --classificator ] arg (=list) Which classificator should be used for
evaluation. Possible: ( grow | none |
one | )
-e [ --evaluation ] arg (=threshold=0.6666)
May be used to override evaluation
settings and default settings from
settings.json
-d [ --dataset ] arg The root path of the evaluation
dataset. The path is expected to
contain features.json, groundtruth.json
and settings.json
This application uses an evaluation dataset to evaluate a specific classificator implementation. The dataset must be formatted as json and can be obtained from group-assignment-datasets.
Different classificators can be implemented and chosen at runtime using the
generic GroupDetector
interface and the corresponding GroupDetectorFactory
.
Currently available classificators are:
A baseline classification that assigns every person to their own group.
A baseline classification that assigns all persons to a single group.
An EM-Based classification that starts from a single group and then alternates between optimizing the group center positions of the current assignment and adding a new group for the person with the highest assignment cost until convergence.
An EM-Based classification that starts with every person in an own group and then alternates between optimizing the group center positions of the current assignment and removing the group with the least 'remove-cost' until convergence.
fformation-gco implements the group assignment using a multi label graph-cuts optimization from [1] as proposed in [2] and the corresponding matlab code (GCFF) using the C++ implementation from gco-v3.0.
[1] Delong A, Osokin A, Isack H. N., Boykov Y (2010) "Fast Approximate Energy Minimization with Label Costs". In CVPR.
[2] Setti F, Russell C, Bassetti C, Cristani M (2015) "F-Formation Detection: Individuating Free-Standing Conversational Groups in Images". In PLoS ONE 10(5): e0123783. doi:10.1371/journal.pone.0123783
GNU LESSER GENERAL PUBLIC LICENSE
This project may be used under the terms of the GNU Lesser General Public License version 3.0 as published by the Free Software Foundation and appearing in the file LICENSE.LGPL included in the packaging of this project. Please review the following information to ensure the license requirements will be met: http://www.gnu.org/licenses/lgpl-3.0.txt