EyeOpenDetection

Training project to detect if eye is open or closed

How to run EyeOpenDetection

Train module

To train multi input model use

python -m train --network mi --dataset_dir /path/to/dataset of the model --output /path/to/output file --image_size image_height image_width

To train single input model use

python -m train --network si --dataset_dir /path/to/dataset of the model --output /path/to/output file --image_size image_height image_width

Where

• --m module type. It can be either train or demo.
• --network network type. It can be either mi or si.
• --output path to model's weight output file.
• --epochs specifies the number of passes through the dataset.
• --batch_size refers to the number of training examples utilised in one iteration usually between 1 and size of the dataset.
• --lr the learning rate of the model.
• --steps how many steps per epoch.
• --weights path to the weights of the model .h5 file.
• --image_size two arguments that specify the height and width of the image.

Dependancies

• tensorflow >= 1.0

• keras >= 2.0

• opencv >= 3.0

• dlib

• numpy

• [shape_predictor_68_face_landmarks.dat][sp]

N.B

• opencv should be compiled with ffmpeg support.
• Conda virtual environment can be created using the following command.

conda env create -f requirements.yml -n emopy_2

Added pretrained model with cropped image size of 24 by 24 which can be used for testing.