I am using DEST in my android face tracking app. The method I use is first detect the face bounding box using DLIB HOG detector (for historic reasons) and then track the face using DEST.
My problem is that it takes DLIB ~50msec to detect the face (with landmarks) but it takes DEST ~160msecs to find landmarks on consecutive frames. Any suggestions as for why? Is the way the bounding box was found affects performance?
BTW, I run dest_track_video on an ubuntu virtualbox and there pedict takes ~200msec.
I am using dest_tracker_vj_ibug.bin tracker file.
FaceTracker.zip
I am using DEST in my android face tracking app. The method I use is first detect the face bounding box using DLIB HOG detector (for historic reasons) and then track the face using DEST. My problem is that it takes DLIB ~50msec to detect the face (with landmarks) but it takes DEST ~160msecs to find landmarks on consecutive frames. Any suggestions as for why? Is the way the bounding box was found affects performance? BTW, I run dest_track_video on an ubuntu virtualbox and there pedict takes ~200msec.
I am using dest_tracker_vj_ibug.bin tracker file. FaceTracker.zip