Biometric, Gender and Age Identification using Mamba

Paper Link - IKT 2024

Overview

This project presents a novel approach to biometric identification that integrates the efficient long-range dependency modeling of Mamba with the U-Net architecture. Our model demonstrates superior accuracy and computational efficiency compared to previous works utilizing transformers and convolutional neural networks (CNNs).

Features

• High Accuracy: Outperforms traditional transformers and CNN-based models in biometric identification tasks.
• Computational Efficiency: Achieves better results with reduced computational overhead.
• Innovative Architecture: Combines the strengths of Mamba for long-range dependency modeling with the U-Net architecture for detailed spatial feature extraction.
• Pretrained Weights: Includes pretrained weights for faster deployment and fine-tuning.

Comparison to Previous Works

Transformers

• Accuracy: Our model achieves higher accuracy rates compared to transformer-based models, particularly in complex biometric datasets.
• Efficiency: Our model is computationally more efficient, reducing the required computational resources and inference time, making it more practical for real-world applications.

CNNs

• Performance: Traditional CNN-based models, while effective, fall short in accuracy when compared to our approach.
• Results: The integration of Mamba's long-range dependency modeling with U-Net’s spatial feature extraction allows the model to capture finer details, leading to better identification results.

Getting Started

Installation

Clone the repository:

git clone https://github.com/Avir-AI/hand_identification_mamba.git
cd ./hand_identification_mamba

Ensure you have Python >= 3.10 installed on your system. Then, install the required libraries and dependencies.

Requirements

pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt

Pre-trained Weights

• Download The Model: best_schedule.pth
• Move best_schedule.pth to: net/pre_trained_weights

Inference (not implemented)

python inference.py --img_path input-path

Training

To train the model, first download the necessary pre-trained weights and datasets:

1. Pretrained Encoder Weights: Download from VMamba GitHub or google drive and move the file to net/pre_trained_weights/vssmsmall_dp03_ckpt_epoch_238.pth.
2. Datasets: Download the dataset of 11k hands images
   • Download datasets
   • unzip and move datasets directory to ./

Run the training process:

python train.py

Testing

After downloading the 11k hands images and , Run the testing process:

python test.py

Acknowledgment

We would like to thank the authors and contributors of SUM for their open-sourced code, which significantly aided this project.

Citation

@inproceedings{rezasoltani2024multi,
  title={A Multi-Task Framework Using Mamba for Identity, Age, and Gender Classification from Hand Images},
  author={Rezasoltani, Amirabbas and Hosseini, Alireza and Toosi, Ramin and Akhaee, Mohammad Ali},
  booktitle={2024 15th International Conference on Information and Knowledge Technology (IKT)},
  pages={41--46},
  year={2024},
  organization={IEEE}
}