NVIDIA GazeNet Facegrid Input

[MikeB_NV]

For Gaze Estimation | NVIDIA NGC, input is three (3) grayscale crops (face, left eye and right) and a face grid. Can you clarify what is meant by a face grid for @bouchemazakary?

[kayccc]

See also:
Questions regarding NVIDIA GazeNet’s input - AI & Data Science / Computer Vision & Image Processing - NVIDIA Developer Forums

[Morganh]

The facegrid is based on fully connected layers. It is a binary face position relative to frame.
See https://arxiv.org/pdf/1606.05814.pdf
The face grid input is a binary mask used to indicate the location and size of the head within the frame (of size 25 × 25) .

[bouchemazakary]

Hi,

I believe that answers my question and also makes total sense. THANK YOU