פרסום פרויקט
התחבר עם פייסבוק
התחבר עם LinkedIn
Entry statistics
0 צפיות
0 צפיות0 שבחים
0 שבחיםEntry content
Entry content
Entry description
This application was developed as a finishing project for Magshimim. The model was developed by us and implemented in Tensorflow. Trained by the dataset named GazeCapture available for free use online.
Here's how my model works:
- We first detect the face using opencv.
- We then take the coordinates of the facebox in relation to the center of the frame.
- We then divife the coordinates of the facebox and the size of the facebox by the height of the frame in pixels to get a normalized set of descriptors about the facebox that won't be affacted by resolution and frame size.
- We then input the facebox itself, and the normalized coordinate into two different branches in our neural network.
- The nerual network looks like this (amount of layers is different) : https://i.imgur.com/RUkY3JH.png
Here's how my model works:
- We first detect the face using opencv.
- We then take the coordinates of the facebox in relation to the center of the frame.
- We then divife the coordinates of the facebox and the size of the facebox by the height of the frame in pixels to get a normalized set of descriptors about the facebox that won't be affacted by resolution and frame size.
- We then input the facebox itself, and the normalized coordinate into two different branches in our neural network.
- The nerual network looks like this (amount of layers is different) : https://i.imgur.com/RUkY3JH.png
הוסף תגובה
0 תגובות
הוסף תגובה
0 תגובות
אנא
היכנס למערכת
כדי להגיב
