How winterwidow/asl-recognition works
This repository is an ASL gesture-recognition application with two main workflows: train an SVM from labeled hand images, and recognize signs from a webcam stream. Both workflows use shared MediaPipe landmark extraction; the live workflow then smooths predictions and displays them. The README also documents PCA visualization and alternate SVM kernels. Training and prediction call sites are excerpted; visualization and dataset capabilities are documented, but their full runtime wiring was not independently inspected.
Main entry points: predict.py Live Predictor · train.py SVM Trainer · utils.py Landmark Extractor · predict_backup.py Image Predictor
Components 13
Model Training
- Gesture Images
data - SVM Trainer
train.py - Number Model
svm_model2.pkl - Alphabet Model
svm_model.pkl - PCA Visualization
train.py
Live Recognition
- Live Predictor
predict.py - Prediction Smoothing
predict.py - Video Overlay
predict.py - Image Predictor
predict_backup.py
Landmark Extraction
- Landmark Extractor
utils.py - Hand Landmarker
hand_landmarker.task
Other components
- Operator
- Webcam
Connections 16
- Operator → SVM Trainer: starts training
README.md - Gesture Images → SVM Trainer: provides images
train.py - SVM Trainer → Landmark Extractor: extracts features
train.py - SVM Trainer → Number Model: saves model
train.py - SVM Trainer → PCA Visualization: plots clusters
README.md - Operator → Live Predictor: starts recognition
README.md - Webcam → Live Predictor: streams frames
predict.py - Live Predictor → Landmark Extractor: extracts features
predict.py - Live Predictor → Alphabet Model: loads model
predict.py - Live Predictor → Prediction Smoothing: buffers predictions
predict.py - Prediction Smoothing → Video Overlay: stabilizes label
predict.py - Live Predictor → Video Overlay: overlays result
predict.py - Image Predictor → Landmark Extractor: extracts features
predict_backup.py - Image Predictor → Number Model: loads model
predict_backup.py - Image Predictor → Video Overlay: shows prediction
predict_backup.py - Landmark Extractor → Hand Landmarker: uses model
utils.py