Abstract In the field of security X ray imaging, existing datasets mainly focus on endogenous domain shifts caused by hardware differences but do not address pseudocoloring driven domain shift (PDS), a type of domain shift resulting from different pseudocoloring schemes across devices. To fill this gap, we construct and release a benchmark dataset designed for studying domain shifts in security X ray inspection, focusing on both PDS and endogenous differences. The dataset has two complementary subsets: a synthetic subset, which applies controlled hue transformations on an endogenous domain shift dataset to isolate and evaluate PDS independently; and an aggregated subset, which integrates multiple public security datasets into a unified format for verifying cross domain and cross device model generalization. All annotations follow standardized structures with detailed descriptions to ensure reproducibility and comparability. This dataset provides a benchmark for evaluating object detection algorithms under domain shifts and supports studies on model robustness and generalization in diverse X ray security scenarios. Similar content being viewed by others Data availability The dataset is available at Figshare29. The PDSXray dataset and analysis code are publicly available at https://doi.org/10.6084/m9.figshare.29958461. The dataset includes X-ray security inspection data for studying the impact of pseudocoloring-driven (PDS) and endogenous (EDS) domain shifts on model performance. The code contains training and analysis scripts, along with model weights, supporting result reproduction. Both the dataset and code are distributed under the Creative Commons Attribution 4.0 International License (CC BY 4.0) (license details: https://creativecommons.org/licenses/by/4.0/). Code availability The code used in this study is hosted on Figshare alongside the PDSXray dataset29: https://doi.org/10.6084/m9.figshare.29958461. It includes scripts for analyzing PDS, cross-dataset category mapping, and model training, along with model weights for replication. The code is licensed under the CC BY 4.0 license, the same as the dataset. The link is public and will remain available after