search

TadasBaltrusaitis / OpenFace

OpenFace – a state-of-the art tool intended for facial landmark detection, head pose estimation, facial action unit recognition, and eye-gaze estimation.
Other
6.98k stars 1.86k forks source link

AU emotion detection #98

Closed ShirleyYim closed 7 years ago

ShirleyYim commented 7 years ago

Hi, I use the Feature Extraction result to predict the facial expression, but I can only detect happy and sad. Could you please give me some advice about how to use the SVM and SVR results together? Thank you very much.

TadasBaltrusaitis commented 7 years ago

Hi,

Your question does not have enough details for me to answer it. In our experiments we found that it is more reliable to recognize some facial expressions from HOG features rather than AUs. So I would start from there rather than necesserilly using AUs directly.

Thanks, Tadas

ShirleyYim commented 7 years ago

Hi, Do you mean that it is more reliable to use HOG features and SVM to directly predict the emotions such as happy/sad? Thanks for your answer.

TadasBaltrusaitis commented 7 years ago

Yes exactly. At least experimentally we found that to be more accurate for categorical emotions.

Thanks, Tadas

ShirleyYim commented 7 years ago

Hi, I trained a svm model using hog features of about 200 facial expression images, and the detection accuracy is very low. Is it because the training dataset too small ? Thanks, Shirley Yim

TadasBaltrusaitis commented 7 years ago

It depends on the complexity of the dataset you used, for some it might be enough for others not necessarily. Furthermore, you need to be careful about feature normalization and might possibly need some dimensionality reduction of the HOG features.

Thanks, Tadas

elahia commented 6 years ago

Hi , What set of outputs, are the HOGs? I mean which columns of 709 are related to HOG?

Thanks

TadasBaltrusaitis commented 6 years ago

Please see: https://github.com/TadasBaltrusaitis/OpenFace/wiki/Output-Format#similarity-aligned-face-images-and-hog-binary-files

HOG are not in the CSV file, but rather in a separate .hog file

elahia commented 6 years ago

Perfect, Thank you,

monajalal commented 5 years ago

@TadasBaltrusaitis can you please provide instruction how to visualize the .hog file for each video segment? How can I layout the .hog file for each aligned frame of a short video?

TadasBaltrusaitis commented 5 years ago

HOG feature visualization is currently done only during the processing of the data, so if you process the original files with -verbose setting you will see the visualization. You can find the code for it in: Visualise_FHOG function in VisualizationUtils.cpp