Models Directory

This directory contains model files used by the hand gesture classification system.

Files

hand_landmarker.task

MediaPipe hand landmark detection model used for feature extraction.

• Purpose: Detects 21 hand landmark points in images
• Used by: FeatureExtractor class
• Source: MediaPipe hand landmarker model
• Download: If missing, manually download from: https://storage.googleapis.com/mediapipe-models/hand_landmarker/hand_landmarker/float16/1/hand_landmarker.task Or see official documentation: https://ai.google.dev/edge/mediapipe/solutions/vision/hand_landmarker/python The system will also attempt to use MediaPipe's bundled model if available.

rps_classifier.joblib

Trained scikit-learn classifier for rock-paper-scissors gesture classification.

• Purpose: Classifies extracted hand features into rock, paper, or scissors
• Created by: Training script (handmotion.train)
• Used by: Prediction script (handmotion.predict)
• Format: Joblib-serialized scikit-learn model

metrics.json

Evaluation metrics from model training.

• Purpose: Stores accuracy, confusion matrix, and classification report
• Created by: Training script (handmotion.train)
• Format: JSON file with evaluation results

Model Pipeline

1. hand_landmarker.task extracts hand landmarks from raw images
2. Features are processed and saved to data/processed/rps/data.npz
3. rps_classifier.joblib is trained on these features
4. Trained classifier is used for prediction on new images