facial-landmarks-35-adas-0002 reference

[KarthickM]

Hi,

I am using smart_classroom_demo in openVINO->inference_engine->demos. With landmarks-regression-retail-0009.xml model, I am able to run the demo application. 

I want to run with facial-landmarks-35-adas-0002.xml to have a more facial landmarks. However, in the code align_transform.cpp file, I see reference landmark values which would work only for 5 facial landmarks (example for model landmarks-regression-retail-0009. Please find the reference values below). I see the same normalized coordinates present in face-reidentification-retail-0095 

static const float h = 112.;
static const float w = 96.;
// reference landmarks points in the unit square [0,1]x[0,1]
static const float ref_landmarks_normalized[] = {
30.2946f / w, 51.6963f / h, 65.5318f / w, 51.5014f / h, 48.0252f / w,
71.7366f / h, 33.5493f / w, 92.3655f / h, 62.7299f / w, 92.2041f / h};

[(0.31556875000000000, 0.4615741071428571),

(0.68262291666666670, 0.4615741071428571),

(0.50026249999999990, 0.6405053571428571),

(0.34947187500000004, 0.8246919642857142),

(0.65343645833333330, 0.8246919642857142)]

Where will I find reference landmark values for facial-landmarks-35-adas-0002? There is already a query for the same here : https://community.intel.com/t5/Intel-Distribution-of-OpenVINO/facial-landmarks-35-adas-0002-reference/m-p/1181229#M18313. But I do not see an answer. 

Please let me know if more information is needed.

Using facial-landmarks-35-adas-0002 will have better results for face reidentification? Currently, I am using landmarks-regression-retail-0009.xml, but the face reidentification results are not good.

Note : I am using C++ version code to run the smart_classroom_demo application.

Warm Regards,

Karthick

[Vladimir_Dudnik]

Hi,

potentially, you may want to look at gaze_estimation_demo which can use the facial-landmarks-35-adas-0002 model to see how key points detected by model are handled.

Regards,
  Vladimir

[IntelSupport]

Hi Karthick,

Facial-landmark-35-adas-0002 model is only validated on Interactive Face Detection Demo and Gaze Estimation Demo. Please refer to the following link: https://docs.openvinotoolkit.org/latest/omz_demos_README.html#demos_that_support_pre_trained_models

On a separate note, I would suggest you explore the available Face Recognition models at the following link: https://docs.openvinotoolkit.org/latest/face_recognition.html

Regards,

Adli

[KarthickM]

Hi Adli,

Thanks a lot. I will check explore those models.

Warm Regards,

Karthick

[IntelSupport]

Hi Karthick,

This thread will no longer be monitored since we have provided a solution. If you need any additional information from Intel, please submit a new question. 

Regards,

Adli