Computer Laboratory

CCNF

Teaser figure
Graphical representation of our CCNF emotion prediction model, x represents the input features and y the continuous output variables with a hidden neural layer in-between.

Summary

An increasing number of computer vision and pattern recognition problems require structured regression techniques. Problems like human pose estimation, unsegmented action recognition, emotion prediction and facial landmark detection have temporal or spatial output dependencies that regular regression techniques do not capture. We present continuous conditional neural fields (CCNF) - a novel structured regression model that can learn non-linear input-output dependencies, and model temporal and spatial output relationships of varying length sequences. We propose two instances of our CCNF framework: Chain-CCNF for time series modelling, and Grid-CCNF for spatial relationship modelling. We evaluate our model on five public datasets spanning three different regression problems: facial landmark detection in the wild, emotion prediction in music and facial action unit recognition. Our CCNF model demonstrates state-of-the-art performance on all of the datasets used.

Relevant publications

CCNF for Continuous Emotion Tracking in Music: Comparison with CCRF and Relative Feature Representation
Vaiva Imbrasaitė Tadas Baltrušaitis, and Peter Robinson
in IEEE International Conference on Multimedia and Expo, Multimedia Affective Computing, Chengdu, China, July 2014
[pdf]

Continuous Conditional Neural Fields for Structured Regression
Tadas Baltrušaitis, Louis-Philippe Morency, and Peter Robinson
in European Conference on Computer Vision, Zürich, Switzerland, September 2014
[pdf]

Constrained Local Neural Fields for robust facial landmark detection in the wild
Tadas Baltrušaitis, Louis-Philippe Morency, and Peter Robinson
in International Conference on Computer Vision, 300 Faces in-the-wild challenge, Sydney, Australia, December 2013
[pdf]

Downloads

Citing our work

If you use any of the resources provided on this page in any of your publications we ask you to cite our work.

Bibtex

@inproceedings{Imbrasaite2014,
  author	=	{Vaiva Imbrasait\.{e} and Tadas Baltru\v{s}aitis and Peter Robinson},
  title		=	{CCNF for Continuous Emotion Tracking in Music: Comparison with CCRF and Relative Feature Representation},
  booktitle	=	{Multimedia Affective Computing, IEEE International Conference on Multimedia and Expo},
  year		=	2014,
}
@inproceedings{Baltrusaitis2014,
  author	=	{Tadas Baltru\v{s}aitis and Louis-Philippe Morency and Peter Robinson},
  title		=	{Continuous Conditional Neural Fields for Structured Regression},
  booktitle	=	{European Conference on Computer Vision},
  year		=	2014,
}
@inproceedings{Baltrusaitis2013,
  author	=	{Tadas Baltru\v{s}aitis and Louis-Philippe Morency and Peter Robinson},
  title		=	{Constrained Local Neural Fields for robust facial landmark detection in the wild},
  booktitle	=	{300 Faces in-the-wild challenge, International Conference on Computer Vision},
  year		=	2013,
}