Sentimental Galaxy

Github: https://github.com/Cara-Cai/Sentimental_Galaxy

Live Demo: https://cara-cai.github.io/Sentimental_Galaxy/

Screen Recording 2023-12-08 at 16.07.32.MOV

Research and Experiments - Facial Recognition

• ml5/Teachable machine/image classification: https://editor.p5js.org/ml5/sketches/ImageModel_TM
• Teachable Machine Model Training Experiments
  • Five buckets don’t work well. The model can’t tell neutral, smile, and laugh when reopened, though it performed well in the teachable machine.

• Retrained with 3 buckets - Neutral / Happy / Sad

  • BAD MODEL!!!!

  

  

  

  • The model performed better when more images were fed
  • Only my images were fed. Not accurate so far. Need more images. And more new faces
  • Alternatives?

• An emotion-recognition-with-ML5 GitHub. I tried this. It's not working well with limited images.

• Question
  • Any trained model I can use?
  • Should I train my model - teachable machine/any other method? How many datasets are expected?
  • Probably through the calculation of key points in the ml5 face mesh?
    • Tried to find trained model - Nah

• My friend suggests: Gestures(?!?!) - easier for TM to be trained

• Oh myyyyyyy - Look what I found!

Real Time Face Detection in p5.js

⬆️This guy saves my life!!!!