Skip to content

installation.md

Latest commit

 

History

History
64 lines (37 loc) · 2.55 KB

installation.md

File metadata and controls

64 lines (37 loc) · 2.55 KB

DryEye Defender

The graphical user interface for the DryEye Defender software: detecting blinks and providing reminders to blink.

demo.png

Repos

Code Structure

For details see code_structure but in short, we this repo contains the GUI implemented in pyside6, and creates a local sqlite3 DB for storing blink data. No personal data is shared to the internet. The module uses a submodule blink-detection to perform inference on webcam frames.

How to install and run

Ensure you have git lfs installed: git lfs install && git lfs pull

Docker

  • To start via docker, simply run docker-compose up
  • For the webcam, the docker-compose mounts a few video sockets in the hope of capturing the correct one for your machine, you may need to edit docker-compose if your webcam is not one of these devices.
  • We mount the current repo so edits to code will be reflected live in the container.

Local

Alternatively, you can run the program locally.

  1. Create a venv of python3.11

    1. python3.11 -m venv .venv
    2. source .venv/bin/activate

  2. Update submodules

    1. git submodule update --init

  3. Install requirement.txt

    1. python -m pip install -r requirements_linux.txt for linux (or requirements_mac.txt / requirements_windows.txt if on Mac or Windows)
    2. for linting and ci libraries : python -m pip install -r requirements_ci.txt

  4. Install the submodule python -m pip install -e submodules/blink-detection 3. or using another solution python -m pip install "git@github.com:scoville/blink-detection-public.gitt"

  5. Run program

    1. python -m dryeye_defender

Building binaries

See the github actions scripts for details on how to build a binary for your platform.

Database

We use a sqlite3 DB to store blink data, for which the Entity-Relationship diagram is shown below:

./ERD.jpg

The blink_history table stores the blink data indexed by timestamp.

The blink_marker column is used to indicate whether the blink was detected by the model or by the user (0 most of the time, 1 if a blink is detected).

blink_value represents the eye state: it is either -1 (closed) or 1 (open)

left_ear and right_ear are floating point numbers representing the Eye Aspect (EAR) Ratio. This is in range -1 to 1 due to normalizing via a rolling window median filter (i.e. centering EAR around 0)