NeuralRot classifies webcam gestures using pose + hand landmarks, then serves live predictions to a web frontend.

1. Data collection:
   • main.py captures webcam frames.
   • MediaPipe extracts pose + left hand + right hand landmarks.
   • Features are normalized and appended to gesture_data.csv.
2. Training:
   • Reads gesture_data.csv.
   • Trains a RandomForestClassifier.
   • Saves model to gesture_model.pkl.
3. Final running:
   • Frontend captures webcam frames.
   • Backend (/predict) extracts the same features and predicts label.
   • Frontend displays the predicted gesture live.

NeuralRot/
|-- main.py
|-- gesture_classifier.py
|-- backend_api.py
|-- meme_engine.py
|-- gesture_data.csv
|-- gesture_model.pkl
|-- requirements.txt
|-- frontend/
|   |-- index.html
|   |-- script.js
|   `-- styles.css
`-- memes/

python -m venv venv
venv\Scripts\activate
python -m pip install -r requirements.txt

Recommended Python: 3.10 to 3.12.

python main.py --labels dab,neutral --samples-per-label 100

python main.py --skip-collect

python gesture_classifier.py --csv gesture_data.csv --model gesture_model.pkl --augment-factor 3

python main.py --run-app

Open:

• http://127.0.0.1:5500

Health check:

• http://127.0.0.1:8000/health

• If predictions do not work, confirm gesture_model.pkl exists.
• If model is missing, train first with python main.py --skip-collect.
• Keep head/shoulders/elbows/hands visible during collection and inference.