Few-shot Concealed Object Detection In Sub-thz Security Images Using Improved Pseudo-annotations

This article discusses the use of fine-tuning based frameworks for few-shot object detection in sub-terahertz security images. The researchers propose an innovative pseudo-annotation method to augment the object detector and enhance its ability to detect challenging objects with limited training examples. They also utilize multiple one-class detectors and a fine-grained classifier trained on thermal-infrared images to prevent overfitting. This approach has been successfully applied for detecting concealed objects in high-contrast passive sub-terahertz imaging.