OpenFace – a state-of-the art tool intended for facial landmark detection, head pose estimation, facial action unit recognition, and eye-gaze estimation.
Other
6.71k
stars
1.82k
forks
source link
Get landmarks corrected for translation, rotation, looming movements. #1025
First of all, thank you for the massive work with this great tool. I have a question that I am hoping it was probably considered during the development of openFace couldn't find an answer from the paper and wiki.
I need to analyze facial landmarks from different videos, such that the analysys is not influenced by the fact that the landmark positions (and distances between landmarks) can change due to movements like head rotation, translation, and looming, which would increase or decrease distances between landmarks. I am interested instead in changes due to expressions only (e.g.facial expressions and speech in the videos).
is there a way to get the landmarks normalized for this, partialing out the variation due to translation, rotation, looming or zooming? I can think of a few ways to correct for this manually, for example by normalising all landmark values based on a fixed distance, of landmark pairs but there are many intermediate choices that it is not obvious what is the best way to approach this.
First of all, thank you for the massive work with this great tool. I have a question that I am hoping it was probably considered during the development of openFace couldn't find an answer from the paper and wiki.
I need to analyze facial landmarks from different videos, such that the analysys is not influenced by the fact that the landmark positions (and distances between landmarks) can change due to movements like head rotation, translation, and looming, which would increase or decrease distances between landmarks. I am interested instead in changes due to expressions only (e.g.facial expressions and speech in the videos).
is there a way to get the landmarks normalized for this, partialing out the variation due to translation, rotation, looming or zooming? I can think of a few ways to correct for this manually, for example by normalising all landmark values based on a fixed distance, of landmark pairs but there are many intermediate choices that it is not obvious what is the best way to approach this.
Thank you!