Open-Vocabulary Prohibited Item Detection for Real-World X-ray Security Inspection

To support the research on the task of open-vocabulary prohibited item detection, we contribute the first X-ray security inspection OVOD evaluation benchmark, named PIXray Caption dataset, which contains 15 common categories and 5,046 image-caption pair annotations. We have released example image-caption pairs from the PIXray Caption dataset. The complete dataset will be released upon the acceptance of the paper.