zhaozz-lab / pupiltracker Goto Github PK

Follow the eye with the webcam. Tracking the pupil with the help of neural networks.
A project made in one month as part of a student internship at the Poznań University of Technology.

• webcam (we use LOGITECH HD PRO C920)
• python >= 3.6.4
• keras >= 2.1.6
• tensorflow/tensorflow-gpu >=1.9
• numpy >= 1.11.3
• opencv-python >= 3.4.1
• dlib >= 19.15.0
• imutils >= 0.4.6

• pupil.py - An application that is designed to detect eyes and save them as a photo. It is needed to create a data set.
• click.py - Designed to indicate pupils and cornea on each photo of the dataset. Photographs and coordinates are normalized and saved to one database.
• Train.ipynb - Convolutional neural network, data loading, preprocessing, training a model
• Train_with_transfer_learning.ipynb - Attempt to use 'transfer learning', 'InceptionV3' algorithm, 'imagenet' weights
• main.py - The main program of the project.

• Damian Szkudlarek
• Bartosz Ptak