mamgai1/Kineticvision

Kinetic Vision — Real-Time Sign Language Translation

A real-time sign language translation app powered by SigLIP (ASL) and a custom LSTM model (ISL), with MediaPipe for hand & pose tracking.

Features

• —ASL Mode — Image-based classification using SigLIP (A–Z)
• —ISL Mode — Keypoint-sequence classification using LSTM (32 classes)
• —Real-time — WebSocket-based inference with live camera feed
• —MediaPipe — Client-side hand & pose landmark detection and visualization

Tech Stack

• —Frontend: Vanilla HTML/CSS/JS + MediaPipe Vision
• —Backend: FastAPI + WebSocket
• —ML Models: SigLIP (HuggingFace Transformers), Custom Keras LSTM
• —Deployment: Docker on HuggingFace Spaces