EmotiW 2016 Challenge

Group Happiness Assessment Using Geometric Features and Dataset Balancing

Here we make available the code employed in our team’s submissions to the 2016 Emotion Recognition in the Wild contest, for the sub-challenge of Group Happiness estimation. The objective of this sub-challenge is to estimate the happiness intensity of a group of people in static images. The HAPPEI dataset used is provided by the organizers.

We followed a predominately bottom-up approach, in which, the individual happiness level of each face is estimated separately, and estimation of group-level happiness is done by either simply averaging all individual estimations, or using a Multi Layer Perceptron. We use geometric features extracted on 49 facial points. Estimation of face-level happiness is cast as a regression problem using Partial Least Squares for the training. Since HAPPEI is a highly skewed dataset (very few training examples for neutral-0 and thrilled-5) various balancing techniques are also explored.

Experimental results (see paper below) show that this regression approach achieves a RMSE of 0.8316 on the competition test set, which compares favorably to the organizers' base line (1.30).

Sample photos

The code (167MB) is released under a creative commons license. Our 4 best-performing models are provided (submissions 1, 3, 4 and 5). The files include both Matlab scripts and data files. The data files include coordinates for all detected faces (using openCV VJ frontal detector), as well as their corresponding facial points, in the whole HAPPEI dataset. The code uses these data in order to train the face-level and group-level happiness intensity models.

The zip-files are encrypted; please email us for the password.

More details about this work can be found in the following paper:

V. Vonikakis, Y. Yazici, V. D. Nguyen, S. Winkler.
Group happiness assessment using geometric features and dataset balancing.
Proc. 18th ACM International Conference on Multimodal Interaction (ICMI), Emotion Recognition in the Wild Challenge, Tokyo, Nov. 12-16, 2016.

Please cite the above paper if you use our code.