askarikzm/hixray-security-xray

HiXray Security X-ray Dataset

Overview

High-quality X-ray security screening dataset used for BYOL (Bootstrap Your Own Latent) self-supervised pretraining.

This dataset is part of an FYP research project on X-ray baggage screening segmentation, in collaboration with Wancom Communications (Pvt) Ltd, Karachi, Pakistan.

Dataset Summary

Split	Images
Train	36,295
Test	9,069
Total	45,364

Image Details

• Format: JPEG
• Content: X-ray scans of passenger bags at security checkpoints
• Objects: Everyday and prohibited items scanned through X-ray machines

Classes (8 categories)

Class	Description
Mobile_Phone	Mobile phones and smartphones
Laptop	Laptop computers
Tablet	Tablet devices
Water	Water bottles and liquid containers
Cosmetic	Cosmetic items
Nonmetallic_Lighter	Non-metallic lighters
Metallic_Lighter	Metallic lighters
Knife	Knives and bladed objects

Usage — BYOL Pretraining

This upload contains images only (no annotations). BYOL is self-supervised — labels are not used during pretraining.

from huggingface_hub import snapshot_download

path = snapshot_download(
    repo_id="askarikzm/hixray-security-xray",
    repo_type="dataset"
)

Training Setup

• Method: BYOL (Bootstrap Your Own Latent)
• Backbone: ResNet-50 (ImageNet pretrained)
• Platform: RunPod GPU cloud
• Epochs: 200
• Framework: PyTorch + byol-pytorch

Citation

If you use this dataset, please cite the original HiXray paper:

@inproceedings{tao2021hixray,
  title={Towards Real-World X-ray Security Inspection},
  author={Tao, Renshuai and others},
  booktitle={ICCV},
  year={2021}
}

License

CC BY-NC 4.0 — Non-commercial research use only.