This project implements a synthetic palmprint image generation system based on Diff-Palm, a diffusion-based model for biometric image synthesis.
The model is inspired by the research paper: https://arxiv.org/abs/2503.18312
The system generates realistic palmprint images to support data augmentation, biometric research, and privacy-preserving applications.
The model is deployed as an interactive web application: https://huggingface.co/spaces/loinguyen5704/diff-palm-app
- Diffusion-based generative model (Diff-Palm)
- High-quality synthetic palmprint generation
- End-to-end ML pipeline
- Hugging Face deployment
- Suitable for biometric applications
-
Input: 1 palmprint image
-
Output: Multiple synthetic variations
-
Preserve main palm lines
-
Simulate different environmental conditions
- Input:
- Number of IDs
- Images per ID
- Output:
- Structured synthetic dataset
Diff-Palm (Diffusion-based Palmprint Generation)
- Data preprocessing
- Diffusion model training
- Image generation
- Visualization
User → Web UI → Inference → Diffusion Model → Output Image
https://huggingface.co/spaces/loinguyen5704/diff-palm-app
├── Diff-Palm.zip # Source or pretrained model package
├── packages.txt # System-level dependencies (for Hugging Face Spaces)
├── README.md # Project documentation
├── requirements.txt # Python dependencies
├── run_diff_palm.sh # Script to run the application
│
├── models/
│└── model.pt # Trained Diff-Palm model weights
│
└── utils/
└── pcem.py # Utility functions (processing / enhancement)
└── app/
└── app.py # Entry point for application (UI / inference)
└── old_approaches/ # Old approaches in Palmprint image classification
- Python
- PyTorch
- Diffusion Models
- OpenCV
- Hugging Face Spaces
pip install -r requirements.txt
python app/app.py
- Add evaluation metrics (FID, SSIM)
- Improve inference speed
- Conditional generation
- Diffusion model implementation
- Research-to-product pipeline
- Real-world deployment
- Nguyen Tran Loi
- Nguyen Nhat Long
- Tran Minh Tam
MIT