Superpowering Open-Vocabulary Object Detectors for X-ray Vision
Pablo Garcia-Fernandez, Lorenzo Vaquero, Mingxuan Liu, Feng Xue, Daniel Cores, Nicu Sebe, Manuel Mucientes, Elisa Ricci
IEEE/CVF International Conference on Computer Vision
RAXO adapts RGB open-vocabulary detectors to X-ray imagery with training-free visual descriptors and introduces DET-COMPASS for large-scale X-ray OvOD evaluation.
Open-vocabulary object detection is set to revolutionize security screening by enabling systems to recognize any item in X-ray scans. However, developing effective open-vocabulary object detectors for X-ray imaging presents unique challenges due to data scarcity and the modality gap that prevents direct adoption of RGB-based solutions. We propose RAXO, a training-free framework that repurposes off-the-shelf RGB open-vocabulary detectors for robust X-ray detection. RAXO builds high-quality X-ray class descriptors using a dual-source retrieval strategy, gathering relevant RGB images from the web and enriching them through an X-ray material transfer mechanism. These visual descriptors replace text-based classification, leveraging intra-modal feature distances for robust detection. Experiments show consistent improvements over base detectors, and the work introduces DET-COMPASS, a benchmark with bounding box annotations for over 300 object categories.
@inproceedings{garciafernandez2025superpowering,
author = {Pablo Garcia-Fernandez and
Lorenzo Vaquero and
Mingxuan Liu and
Feng Xue and
Daniel Cores and
Nicu Sebe and
Manuel Mucientes and
Elisa Ricci},
title = {Superpowering Open-Vocabulary Object Detectors for X-ray
Vision},
booktitle = {{IEEE/CVF} Int. Conf. Comput. Vis. ({ICCV})},
pages = {20770--20779},
year = {2025}
}