Dashboard Overview
Real-time status, labels inventory, and models overview.
0
Total Signs Registered
0
Sequence Samples
Checking...
LSTM Model Status
~30 FPS
Camera Performance
Welcome to ScribeSign AI
This is a high-performance **MediaPipe Holistic** and **LSTM Neural Network** computer vision engine designed to translate complex sign language motions (incorporating face mesh, body pose, and both hands) directly into coherent English speech and sentences.
Quick Actions
Registered Signs Inventory
0 Signs| Sign / Gloss Label | Dataset Samples | Action |
|---|---|---|
| Loading dataset... | ||
System Status & Logs
Active Gloss Buffer
Buffer of stable detected gloss keywords (max 5) fed to the NLP Sentence Reconstructor:
Reconstruction Logs
Record Landmark Sequences
Enter a custom word/label. Pressing **Record** will start a 3-second visual countdown and capture exactly **30 frames** of holistic landmarks.
Recording Guidelines
- Ensure your entire face, shoulders, and both hands are clearly visible in the camera frame.
- Perform the gesture continuously throughout the recording period.
- Try varying camera angles and hand positions slightly in each sequence to achieve a highly robust model.
- Aim for at least **15-20 sequences per sign** for reliable accuracy.
Holistic Dataset Inventory
Current gesture directories and sequences saved in `Data/Holistic_Landmarks`:
LSTM Training Dashboard
Fine-tune training options and trigger network training over collected landmarks. This process will disable real-time inference temporarily.
Real-time Training Metrics Plot
Training System stdout
consoleLandmarks Files Browser
Select a recorded `.npy` sequence file from the folders below to play it back in the skeletal viewer.