How iamafrar/ai-based-sinhala-sign-language-translator works
SSLR is a real-time Sinhala Sign Language recognizer: the user starts a webcam, the frontend sends frames to FastAPI, and the backend accumulates a sequence, extracts pose landmarks, scales the features, predicts a word, and returns it for display. The README also documents a separate dataset-preparation, model-training, and evaluation workflow in the notebook. Those notebook stages are included as documented capabilities; their internal implementation and connections to deployed assets were not inspected.
Main entry points: backend_fastapi.py FastAPI endpoints · app.js Camera capture · index.html Web interface · SSL_WORD_LEVEL_80_PLUS.ipynb Model training
Components 15
Web client
- Web interface
frontend/index.html - Camera capture
frontend/app.js - Prediction display
frontend/app.js
Prediction service
- FastAPI endpoints
backend_fastapi.py - Frame decoding
backend_fastapi.py - Pose landmarks
backend_fastapi.py - Sequence buffer
backend_fastapi.py - Feature scaling
backend_fastapi.py - BiLSTM attention
backend_fastapi.py - Word decoding
backend_fastapi.py
Model development
- Data preparation
SSL_WORD_LEVEL_80_PLUS.ipynb - Model training
SSL_WORD_LEVEL_80_PLUS.ipynb - Model evaluation
SSL_WORD_LEVEL_80_PLUS.ipynb
Other components
- User
- Webcam
Connections 14
- User → Web interface: uses
README.md - Webcam → Camera capture: provides frames
README.md - Web interface → Camera capture: hosts
README.md - Camera capture → FastAPI endpoints: posts frames
frontend/app.js - Camera capture → FastAPI endpoints: resets sequence
frontend/app.js - FastAPI endpoints → Frame decoding: decodes upload
backend_fastapi.py - Frame decoding → Pose landmarks: extracts pose
backend_fastapi.py - FastAPI endpoints → Sequence buffer: buffers features
backend_fastapi.py - FastAPI endpoints → Feature scaling: scales sequence
backend_fastapi.py - FastAPI endpoints → BiLSTM attention: predicts
backend_fastapi.py - FastAPI endpoints → Word decoding: decodes classes
backend_fastapi.py - FastAPI endpoints → Prediction display: returns prediction
frontend/app.js - Data preparation → Model training: prepares data
README.md - Model training → Model evaluation: evaluates model
README.md