This page introduces the SIXray-D annotations for the task of detection of prohibited items in the SIXray dataset. For further details, please see our paper: https://www.mdpi.com/2072-666X/13/4/565
The SIXray dataset contains X-ray images collected from security scanners at several subway stations. There are six categories of prohibited items, namely, gun, knife, wrench, pliers, scissors, and hammer. The annotations provided by this dataset are for classification purposes, but do not provide localization information of prohibited items. The SIXray dataset can be obtained from https://github.com/MeioJane/SIXray
- SIXray: The original SIXray dataset. It has positive folders that have positive images containing prohibited items and 20 negative folders (folder 0~19) of negative images.
- Positive images: images with prohibited objects (Gun, Knife, Scissors, Wrench, Pliers, Hammer).
- Negative images: images without prohibited objects.
- SIXray-D: The SIXray-D annotations contains the annotations for the detection task. Only 5 classes of prohibited items are annotated, namely, gun, knife, wrench, pliers, and scissors.
We conducted thorough inspection on the original SIXray dataset and annotated all the positive objects with bounding boxes. The false-negative and false-positive images in the original dataset are also detected and annotated. Some observations can be made:
- Many images are different viewing angles of the same objects.
- Many images (especially long images) contain white background.
- Image that are identical but with different aspect ratio (with white background) are commonly seen in the dataset.
- Many images contain multiple prohibited items.
- Many images contain multiple bags.
- Many images contain the same object with and without cropping of the white background.
Release data statistics:
Number of positive images: 11,401
Number of annotation boxes: 23,470
|No. of annotation boxes||3,930||3,237||5,048||7,070||4,185||23,470|
|Percentage of annotation boxes per class||16.7%||13.8%||21.5%||30.1%||17.8%||100%|
Number of annotation boxes sorted by aspect ratio (AR):
|Knife||5 (0.1%)||1,079 (33.3%)||763 (23.6%)||470 (14.5%)||320 (9.9%)||600 (18.5%)||3,237 (100%)|
|Scissors||18 (0.5%)||2,683 (68.3%)||993 (25.3%)||162 (4.1%)||39 (1.0%)||35 (0.9%)||3,930 (100%)|
|Pliers||33 (0.5%)||4,012 (56.7%)||1,946 (27.5%)||813 (11.5%)||199 (2.8%)||67 (0.9%)||7,070 (100%)|
|Wrench||12 (0.3%)||1,751 (42.8%)||1,070 (25.6%)||694 (16.6%)||348 (8.3%)||310 (7.4%)||4,185 (100%)|
|Gun||24 (0.5%)||3,755 (74.4%)||903 (17.9%)||253 (5.0%)||80 (1.6%)||33 (0.7%)||5,048 (100%)|
|Total||92 (0.4%)||13,280 (56.6%)||5,675 (24.2%)||2,392 (10.2%)||986 (4.2%)||1,045 (4.5%)||23,470 (100%)|
Sample images (with annotation boxes drawn):
The sample images below come from the SIXray dataset, with the bounding box annotations from SIXray-D.
How to obtain the Annotations:
To download the SIXray-D annotations, please see the request access form at the bottom of this page.
Usage for Academic Research
The SIXray-D annotations is released for academic research only and is free to researchers from educational or research institutes for non-commercial purposes.
The use of the SIXray-D annotations is governed by the following terms and conditions:
- Without the expressed permission of the ROSE Lab, any of the following will be considered illegal: redistribution, derivation or generation of a new annotation from this annotation, and commercial usage of any of these annotations in any way or form, either partially or in its entirety.
- All users of this annotation agree to indemnify, defend and hold harmless, the ROSE Lab and its officers, employees, and agents, individually and collectively, from any and all losses, expenses, and damages.
All publications using the SIXray-D annotations should include the following acknowledgement: “(Portions of) the research in this paper used the SIXray-D annotations made available by the ROSE Lab at Nanyang Technological University, Singapore.”
Any publications that results from using the SIXray-D annotations should cite the following paper:
- H. D. Nguyen, R. Cai, H. Zhao, A. C. Kot, and B. Wen, “Towards More Efficient Security Inspection via Deep Learning: A Task-Driven X-ray Image Cropping Scheme,” Micromachines, vol. 13, no. 4, p. 565, Mar. 2022, doi: 10.3390/mi13040565.
If interested, please click on the “Request Annotation” hyperlink below for a copy of the Release Agreement. With your acceptance of the agreement, we will then send you the LoginID and password to the annotations.