search

TadasBaltrusaitis / OpenFace

OpenFace – a state-of-the art tool intended for facial landmark detection, head pose estimation, facial action unit recognition, and eye-gaze estimation.
Other
6.73k stars 1.83k forks source link

Issues about ce-clm_training #728

Open Nourhan27 opened 5 years ago

Nourhan27 commented 5 years ago

Hello,

For CEN patch detector training we have to train a model for every landmark , scale , view, so if i use 2 views and 4 scales, we will have 8 combinations for training, right ??

1- Should we do all training combinations ? 2- How do we integrate the generated models of different landmarks to generate these final models ready to use in C++ detection model ---->> cen_patches_0.25_of.dat cen_patches_0.35_of.dat cen_patches_0.50_of.dat cen_patches_1.00_of.dat

Thanks,

TadasBaltrusaitis commented 4 years ago

Hi,

Apologies for the delayed response.

You're right that fore every landmark, scale and view we need to train each landmark (this can be reduced a bit by using mirrored views, e.g. for left profile and right profile the patch experts can be mirrored and don't need to be retrained).

To generate models for C++ detection, first need to convert the keras models to matlab ones using keras2matlab.py, followed by instructions in matlab_version\models\cen\readme.txt.