sehrish-jabeen/Sign-Language-Production-GCDM

Sign Language Production and Translation Pipeline (GCDM & NSLT Ensemble)

This project features an end-to-end pipeline for Sign Language:

1. SLP (Text-to-Video): Uses a Conditional Diffusion Model (GCDM) to generate 3D skeleton videos from text.
2. SLT (Video-to-Text): Uses an NSLT Model for Back-Translation evaluation.

Results

• Pose Generation: MAE 0.1296, DTW-P 2.7499
• Translation (Real Data): BLEU 7.22, ROUGE 27.63
• Translation (Synthetic Data Adaptation): BLEU 1.37 (Mixed Training)

Quick Start

1. Install dependencies: pip install torch torchvision pandas numpy scipy sacrebleu rouge-score jiwer gradio opencv-python
2. Run the interactive app: python app.py