I am using shape_predictor_81_face_landmarks.dat in order to obtain facial landmarks. Here is the result:

As you you can see this is the points are actually 2D numpy arrays:

[[201 122]
 [201 140]
 [202 157]
 [204 174]
 [209 190]
 [219 205]
 [230 219]
 [244 231]
 [261 236]
 [278 233]
 [292 222]
 [304 209]
 [314 195]
 [320 179]
 [324 163]
 [325 147]
 [327 130]
 [214 119]
 [223 112]
 [236 111]
 [248 113]
 [260 118]
 [277 119]
 [289 116]
 [301 116]
 [312 119]
 [319 127]
 [268 127]
 [267 138]
 [267 148]
 [267 160]
 [250 164]
 [258 167]
 [266 170]
 [274 168]
 [281 165]
 [228 125]
 [235 122]
 [243 123]
 [251 128]
 [242 127]
 [235 127]
 [283 131]
 [291 127]
 [298 127]
 [305 131]
 [298 132]
 [291 131]
 [231 178]
 [243 177]
 [256 178]
 [265 180]
 [274 179]
 [286 180]
 [296 182]
 [285 196]
 [273 202]
 [263 203]
 [254 201]
 [241 193]
 [235 179]
 [255 181]
 [264 183]
 [274 182]
 [293 183]
 [273 195]
 [264 196]
 [255 194]
 [219  73]
 [232  73]
 [252  78]
 [274  80]
 [305  78]
 [318  82]
 [328 108]
 [209  91]
 [216  79]
 [201 114]
 [328 124]
 [321  92]
 [297  81]]

From such data, I want to be able predict people personnality according to some pseudo-science labeling. How should I approach this problem? Should I train some model? Or just go with some raw algorithmes.

>Solution :

As far a s I understand from the question you want to predict the personality of the people in the image. I don’t know the exact application you want but you can use the big five personality dimensions as personality classes. Afterwards, you can use a Face – Personality dataset such as The Basel Face Database (BFD) to train a NN in order to make personality predictions from face images.