AU detector (TAUD 2011)

TAUD (Temporal-based Action Unit Detection) is the implementation of our LPQ-TOP-based AU detector. It is developed as a WIN32 executable. It includes trained models for the following AUs: AU1, AU2, AU4, AU5, AU6, AU7, AU10, AU11, AU12, AU14, AU15, AU17, AU18, AU20, AU22, AU24, AU25, AU26, AU27, AU45 and AU46. The model is trained on the MMI database. The input can either be a video sequence (.avi) or a set of images that form a video (.png). The system assumes that the image/video contains a frontal view of the face and that the video frame rate is 25 fps. Downsampling will be required for videos with a different frame rate. The detector outputs a .csv file with its prediction for each frame (1 for activate, otherwise -1) and saves it in the same directory as that containing the images or the video. Note that TAUD is unable to detect the AU activation in the first and last couple of frames (the descriptors used are window-based, so for example when the window is of 11 frames it will not be possible to detect the activation of the AU for the first and last 5 frames). This is a first version of the software, further versions are expected in the near future.

TOP is the implementation of our TOP-based feature extraction (LBP-TOP and LPQ-TOP). It extracts LBP-TOP or LPQ-TOP features from the images in a directory or from a video using a sliding window. It saves the results in the same directory as that containing the images or the video.

More information about it can be found in:  Bihan Jiang, Michel F. Valstar, and Maja Pantic, 'Action Unit detection using sparse appearance descriptors in space-time video volumes', IEEE Int'l. Conf. Face and Gesture Recognition (FG'11), March 2011.

We kindly request that you cite this work if you decide to use this for research purposes.

You will need Matlab Component Runtime (MCR) version 7.14 installed on your computer for the program to run. We provide the executable to do this. MCR are version dependent. The code is compiled using version 7.14. If you have Matlab 2010b, then you already have it installed. Otherwise, keep in mind you might need to install it in order to run the executables provided here. You can safely install different versions of MCR in your computer.

Executables: