Skip to first unread message
noname8563
Jun 3, 2017, 6:00:27 AM6/3/17
to CMU-OpenFace
I use OpenFace to extract features from a large dataset for further classification. I have the two question which relate to the same problem (accuracy improving):
1 How should I clean my dataset and should I? There are many faces which -are not clear; -expose some grimace; -rotated to a big angle from being straight to camera. I suppose that they result in false features and removing them would improve accuracy. Is there any confidence indicator in the library itself or should use some external tool?
2. I'm especially curious about faces that are rotated around vertical axis, so that one cheek is larger than the other. The thing is that most faces are not exactly full-face but rotated to some extent. How much OpenFace resistant to this?
1 How should I clean my dataset and should I? There are many faces which -are not clear; -expose some grimace; -rotated to a big angle from being straight to camera. I suppose that they result in false features and removing them would improve accuracy. Is there any confidence indicator in the library itself or should use some external tool?
2. I'm especially curious about faces that are rotated around vertical axis, so that one cheek is larger than the other. The thing is that most faces are not exactly full-face but rotated to some extent. How much OpenFace resistant to this?
Reply all
Reply to author
Forward
0 new messages