Generic object detection via image style transfer technique for x-ray baggage scanning devices
Büşra Küçükateş Yalçın, Duygu Selin Ak, Şükrücan Taylan Işıkoğlu · 2025
X-Ray baggage scanning devices vary widely in their features and coloring algorithms, leading to challenges in applying object detection models across different devices due to mismatched coloring schemes. This requires large device-specific datasets, significantly increasing the workload for training and testing models. Moreover, x-ray images from one device are often unsuitable for others. To address these issues, we propose using style transfer techniques to adjust the semantic texture and style of x-ray images according to a reference style. This approach allows for the adaptation of images to meet the requirements of various devices, reducing the need for extensive dataset collection and streamlining the training process for object detection models across different x-ray scanning technologies.