You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Access Request — HUTECH-CNPV1549 Palm Vein Dataset

This dataset contains near-infrared (NIR) palm vein images collected at Ho Chi Minh City University of Technology (HUTECH), Vietnam. Access is restricted to non-commercial academic research only.
By submitting this form, you confirm that:
1. The dataset will be used solely for non-commercial academic research.
2. You will not redistribute or share the dataset with third parties.
3. Any publication using this dataset must cite the original paper.
4. You will delete the dataset upon completion of your research project.

Fill in all fields completely — incomplete requests will be declined.

Log in or Sign Up to review the conditions and access this dataset content.

HUTECH-CNPV1549 Palm Vein Dataset

A contactless near-infrared (NIR) palm vein dataset collected at Ho Chi Minh City University of Technology (HUTECH), Vietnam.

Dataset Summary

Property Value
Subjects 1,549
Images per subject 10
Total images 15,490
Modality Near-infrared (NIR) palm vein
Capture type Contactless
Collection site HUTECH University, Ho Chi Minh City, Vietnam

Versions

Version Path Resolution Description
ROI HUTECH_CNPV1549_ROI/ 128 × 128 px Perspective-warped palm ROI, grayscale BMP
Full HUTECH_CNPV1549_FULL/ 640 × 480 px Full-frame NIR image, grayscale BMP

Data Structure

HUTECH_CNPV1549_ROI/          # or HUTECH_CNPV1549_FULL/
├── identity_0001/
│   ├── img_01.bmp
│   ├── img_02.bmp
│   └── ... (10 images)
├── identity_0002/
│   └── ...
└── identity_1549/
    └── ...

Identity labels run from identity_0001 to identity_1549.

Official Train / Test Split

The file split_openset.json defines the open-set identity-disjoint split used in the benchmark:

  • Train: 1,084 identities
  • Test: 462 identities (no overlap with train)

Protocol: open-set verification — test identities are unseen during training.

Benchmark Results (SCA-MobileNet)

Evaluated under open-set identity-disjoint protocol (70 / 30 split at the identity level).

Protocol EER (%) ↓ TAR @ 0.01% FAR ↑ TAR @ 0.1% FAR ↑ AUC ↑
In-domain 0.89 96.93% 98.49% 0.9983
Cross-domain → TONGJI 1.50 71.84% 94.46% 0.9981
Cross-domain → VERA 3.60 89.25% 90.53% 0.9887

Access

This dataset requires manual approval. Submit a request using the form above.

Requests with a clear institutional affiliation and research purpose are typically processed within 3–5 business days.

Citation

If you use this dataset, please cite:

@article{scamobilenet2026,
  author    = {Bui, Danh Huong and Hoang, Van Quy and Nguyen, Quoc Vuong and Huynh, Minh Quan},
  title     = {{SCA-MobileNet}: A Lightweight Architecture for Contactless Palm Vein Verification},
  year      = {2026}
}

License

CC BY-NC 4.0 — Free for non-commercial academic research only.

Downloads last month
10