1,059,231 X-ray images collected from subway security checkpoints, categorized into 6 prohibited item classes including guns, knives, and hammers. The dataset specifically addresses the extreme class imbalance found in security screening by providing a vast majority of negative samples alongside positive instances.
Use Cases
- Train object detection algorithms to locate illegal items using the provided bounding box coordinates
- Benchmark classification models on the 6 prohibited item labels to improve automated security screening
- Research long-tail distribution and class imbalance techniques using the high ratio of negative-to-positive samples
Strengths
- 1,059,231 total X-ray images sourced from real-world subway security checks
- 6 prohibited item categories: Gun, Knife, Wrench, Pliers, Scissors, and Hammer
- Includes bounding box annotations for all positive instances of prohibited items
- Features three distinct subsets (SIXray10, SIXray100, SIXray1000) to evaluate model performance under varying class imbalance ratios