Googly Eyes

Make faces look funnier!

Description • Getting Started • System Details • Results • Make It More Fun • References

Description

Googly Eyes is a fun project that adds a touch of humour to faces by making them look funnier with googly eyes.

Getting Started

Run via Docker

• Install docker
• Navigate to the project directory: cd googly-eyes
• Run application:

docker-compose up --build

• And go to http://localhost:8501

Run via Setting Up Dev Environment

• Install poetry
• Navigate to the project directory: cd googly-eyes
• Install necessary dependencies:

poetry config virtualenvs.in-project true                  
poetry install 

• Run the backend server

poetry run uvicorn googly_eyes.backend.server:app --reload   

• On new terminal, run the frontend viewer:

cd googly-eyes/googly_eyes/frontend
poetry run streamlit run view.py

• And go to http://localhost:8501

FastAPI docs

Go to http://localhost:8000/docs to see and try available backend api

Run via Python Script

• Install poetry
• Navigate to the project directory: cd googly-eyes
• Install necessary dependencies:

poetry config virtualenvs.in-project true                  
poetry install 

• Run example python script:

poetry run python example.py <path-to-image-file>

System Details

High-Level Architecture

Backend

The backend system handles the googlification of eyes with the following steps:

• Detects faces using BlazeFace.
• Extracts facial landmarks using FaceMesh for all detected faces.
• Googlify eyes
  • Extracts eyes - eye centers and eye sizes; this is randomized to make it fun.
  • Adds a googly eyes image filter based on the extracted eye centers and sizes, with random rotation for added humor.

Frontend

The frontend is a Streamlit viewer providing an interactive interface for users to enjoy the googly eyes effect on faces.

Results

Make It More Fun

There's a config file available! Check out config.yaml and feel free to experiment with the variables. These variables have specific roles:

• Variables under face_detection_settings:
  • score_threshold - face detection confidence score between 0 and 1. A higher value may result in more accurate but fewer detections.
  • iou_threshold - intersection over union threshold for face detection to balance precision and recall, value between 0 and 1.
• Variables under googly_eye_settings:
  • path - image path to googly eye image filter
  • size_multiplier- value to scale to get bigger googly eyes
  • size_inc_percent - to control the maximum random increase in each googly eye's size
  • centre_offset_percent - to change the offset of the googly eyes' centres, makes the googly eyes placement more dynamic.

References

1. BlazeFace Paper
2. FaceMesh Paper
3. MediaPipe

---

author: Shubham M Wagh