This project contains the code for my master thesis.
It revolves around the Machine Learning model to measure the shrinkage of the pupil diameter during a spontaneous light impulse on mobile devices.
How to set up the environment: * $conda create --name thesis_kamphake * $conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch (depends on cuda version) * $cd yolov5 * $pip install -r requirements.txt * $pip install wandb (optional)
How create the Dataset:
* Download the Unity Model from https://www.cl.cam.ac.uk/research/rainbow/projects/unityeyes/
* Generate images with the prefered settings (1600x1200, fantastic, 0-15, 0-15)
* Move the images to pupil_detection/raw_dataset/imgs/
* Start jupyter-notebook
* Open UnityEyes_to_Dataset
* Change if False: to if True
* Follow the instructions in the Notebook
How to train the models: * Prepare the Dataset * Create a Weights and Biases account for tracking (highly encouraged) * Follow the instructions in the notebook