SIXray : A Large-scale Security Inspection X-ray Benchmark for Prohibited Item Discovery in Overlapping Images (1901.00303v1)

Abstract: In this paper, we present a large-scale dataset and establish a baseline for prohibited item discovery in Security Inspection X-ray images. Our dataset, named SIXray, consists of 1,059,231 X-ray images, in which 6 classes of 8,929 prohibited items are manually annotated. It raises a brand new challenge of overlapping image data, meanwhile shares the same properties with existing datasets, including complex yet meaningless contexts and class imbalance. We propose an approach named class-balanced hierarchical refinement (CHR) to deal with these difficulties. CHR assumes that each input image is sampled from a mixture distribution, and that deep networks require an iterative process to infer image contents accurately. To accelerate, we insert reversed connections to different network backbones, delivering high-level visual cues to assist mid-level features. In addition, a class-balanced loss function is designed to maximally alleviate the noise introduced by easy negative samples. We evaluate CHR on SIXray with different ratios of positive/negative samples. Compared to the baselines, CHR enjoys a better ability of discriminating objects especially using mid-level features, which offers the possibility of using a weakly-supervised approach towards accurate object localization. In particular, the advantage of CHR is more significant in the scenarios with fewer positive training samples, which demonstrates its potential application in real-world security inspection.

• The paper introduces SIXray, a large-scale security inspection X-ray benchmark dataset with over a million images, designed to address the challenge of detecting prohibited items in complex, overlapping images.
• To handle the data's complexity and class imbalance, the authors propose the Class-Balanced Hierarchical Refinement (CHR) strategy, which iteratively analyzes layered images and uses a class-balanced loss to enhance detection performance.
• This research has significant practical implications for automating security inspections in high-density areas and theoretical implications for advancing hierarchical network architectures and handling sample imbalance in deep learning.

An Overview of SIXray: A Benchmark for Prohibited Item Discovery in Security Inspection X-ray Images

The research paper "SIXray: A Large-scale Security Inspection X-ray Benchmark for Prohibited Item Discovery in Overlapping Images" addresses the critical challenge of detecting prohibited items in security inspection X-ray images, which are essential for maintaining safety in public spaces. The authors present a newly curated dataset, SIXray, comprising over one million images, annotated to identify six categories of prohibited items with a particular emphasis on items such as guns, knives, and scissors.

Key Contributions and Methodologies

1. Dataset Significance and Properties:

SIXray constitutes a substantial leap in available data for X-ray inspection tasks, exceeding previous datasets like GDXray by over one hundred times in size. Noteworthy is the real-world complexity reflected in SIXray, characterized by the overlapping nature of objects, varied scales and viewpoints, and a high imbalance between positive images (those containing prohibited items) and negatives.

2. Class-Balanced Hierarchical Refinement (CHR):

To navigate the challenges presented by duplicative backgrounds and class imbalances, the authors introduce the Class-Balanced Hierarchical Refinement (CHR) strategy. This approach assumes that each X-ray image is an overview of multiple layered images and employs deep networks with reversed connections to iteratively analyze the data. By doing so, CHR enhances performance by focusing on mid-level features for improved object localization and detection. It also introduces a class-balanced loss mechanism to manage negative sample noise, significantly beneficial in scenarios with few positive training images.

3. Performance Evaluation:

Experiments were conducted using CHR on several subsets with varying ratios of positive to negative samples, showcasing its superiority over baseline models, particularly when training data is heavily skewed towards negative samples. CHR's ability to leverage weakly-supervised data for object localization was demonstrated through its application in identifying prohibited items amidst a complex and cluttered backdrop.

Implications and Future Prospects

This research holds substantial practical implications, especially in automating security inspections in high-density environments such as airports and subway stations. The introduction of SIXray paves the way for further advancements in X-ray detection technology, encouraging innovations leveraging deep learning to handle complex, real-world image data.

Theoretically, the paper ignites discussions on improving hierarchical network architectures and developing loss functions that compensate for sample imbalance. Moving forward, exploration into the precise modeling of overlapping images and their extensions to natural image domains could spearhead developments in broader applications of computer vision and machine learning.

In conclusion, this paper lays down a critical foundation for enhancing security inspection mechanisms through advanced machine learning techniques, promising improvements not only in accuracy but also in operational efficiency and reliability.