Dataset and Baseline Method for Long-range X-ray Image Enhancement, 3D Reconstruction, and Object Recognition

Written by: Riley C. W. O’Neill et al.
Published on:

Summary

We consider x-ray imaging over long distances, e.g. 1km+ separation between the source and receiver. Several challenges stand in the way of long-range X-ray image understanding, including sparse photon arrival (resulting in low signal-to-noise ratios), difficulty in source/detector placement (resulting in few views), and motion of the sensor and/or object (resulting in blur). It’s unclear at what distance these challenges preclude effective x-ray image understanding. To address this, we construct a dataset via physics-based simulation to advance the development of six tasks in long-range X-ray scanning: low-flux x-ray image enhancement, 3D reconstruction, novel view rendering, object classification, materials classification, and 3D scene segmentation. The purpose of this dataset is to determine the limits of image understanding with respect to low flux x-ray imaging, and to motivate the development of new x-ray inspection systems with large source-detector separation. Experiments show that conventional x-ray tomographic reconstruction – including recent machine learning methods – fail to produce useful outputs from very low flux imagery. This demonstrates the necessity of developing new low-flux approaches, so we present a physically informed NeRF model in an analysis-by-synthesis framework that serves as a baseline method. Our dataset and source code will be made public.

Publication

O’Neill, R. C. W., Smith, E., Coleman, S. J., Zukić, D., & McCloskey, S. (2026). Dataset and Baseline Method for Long-range X-ray Image Enhancement, 3D Reconstruction, and Object Recognition. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW 2026), 7298-7308.

Citation

@inproceedings{O'Neill_2026_CVPR,
  author    = {O'Neill, Riley C W and Smith, Eric and Coleman, Stephen J and Zukic, Dzenan and McCloskey, Scott},
  title     = {Dataset and Baseline Method for Long-range {X-ray} Image Enhancement, {3D} Reconstruction, and Object Recognition},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
  month     = Jun,
  year      = 2026,
  pages     = {7298-7308},
  url = {https://openaccess.thecvf.com/content/CVPR2026W/VISION26/papers/
    ONeill_Dataset_and_Baseline_Method_for_Long-range_X-ray_Image_Enhancement_3D_CVPRW_2026_paper.pdf},
}