This is an alpha version of the nod shake detector and 5 dimensional emotion predictor trained on Semaine data. The implementation was carried out by Hatice Gunes. The code has been compiled on a Windows machine and it is provided here as a Microsoft Installer file (NSDetector5DEmotionPredictor.msi).
The detector uses the Viola & Jones face detector and works fine for most near-frontal-view faces. The detector-predictor takes an .avi video file as input, detects the nods and shakes occurring in the video by using a processing window of 20 video frames, and outputs predictions in 5 emotional dimensions (arousal, expectation, intensity, power, and valence) in the scale of 0 to +1. The predictors have been trained using the Semaine database and the dimensional annotations provided by 3 human raters as ground truth. The final output for each dimension is an unweighted average of the 3 models (based on the rater models).
More information about the detector-predictor can be found in:
H. Gunes and M. Pantic, "Dimensional Emotion Prediction from Spontaneous Head Gestures for Interaction with Sensitive Artificial Listeners", in Proc. Int'l Conf. on Intelligent Virtual Agents (IVA'10), pp. 371-377, 2010.
We kindly request you to cite this work if you decide to use this tool for research purposes.