How arjunraj-01/sign-language works
SignBridge is a webcam-based ASL alphabet recognizer that turns detected gestures into text. Its main user workflow runs from camera frames through hand detection, image classification and landmark-based disambiguation, then stabilizes predictions and lets users build words with controls. The repository also includes a separate dataset preparation and TensorFlow training/evaluation path. The sampled inference code shows detector and classifier behavior; some app smoothing details and several documented training capabilities are not fully shown, so uncertain links are omitted or marked unverified.
Main entry points: app.py Streamlit app · train.py Model training · inference.py Inference engine · preprocessing.py Dataset loading
Components 15
User interaction
- Streamlit app
app.py - Word builder
src/word_builder.py - Prediction smoother
src/smoothing.py
Live recognition
- Inference engine
src/inference.py - Hand detector
src/hand_detection.py - Landmark disambiguation
src/landmark_rules.py
Model preparation
- Dataset loading
src/preprocessing.py - CNN model builders
src/model.py - Model training
train.py - Model evaluation
evaluate.py
Shared settings
- Runtime configuration
src/config.py
Other components
- Signer
- Webcam
- ASL image dataset
- Trained classifier
Connections 17
- Signer → Streamlit app: operates
README.md - Webcam → Streamlit app: provides frames
README.md - Streamlit app → Inference engine: requests predictions
app.py - Inference engine → Hand detector: detects hand
src/inference.py - Inference engine → Landmark disambiguation: disambiguates
src/inference.py - Inference engine → Prediction smoother: feeds predictions
- Streamlit app → Word builder: updates text
app.py - Streamlit app → Prediction smoother: stabilizes output
app.py - ASL image dataset → Dataset loading: loads images
src/preprocessing.py - Model training → Dataset loading: loads splits
train.py - Model training → CNN model builders: builds classifier
train.py - Model training → Trained classifier: saves weights
train.py - Model evaluation → Dataset loading: loads data
evaluate.py - Model evaluation → Trained classifier: evaluates
- Model training → Runtime configuration: reads settings
train.py - Dataset loading → Runtime configuration: reads settings
src/preprocessing.py - Streamlit app → Runtime configuration: uses defaults