Stacked Boxes—Detection & Segmentation

Accurately detecting and segmenting stacked boxes in images or video frames is a critical task in various domains, such as warehousing, logistics, and robotics. This paper presents a novel approach that combines the You Only Look Once (YOLO) object detection framework with the Segment Anything Model (SAM), a state-of-the-art segmentation model. Our method efficiently detects the bounding boxes of stacked boxes in input images or video frames using the YOLO framework. Then it utilizes the SAM model to perform pixel-level segmentation of the detected bounding boxes. By combining these two approaches, our proposed method achieves precise detection and segmentation of stacked boxes, providing essential information about their location, size, and shape. This information can be leveraged for diverse applications, including inventory management, automated stacking systems, and robotic manipulation tasks. Our experimental results demonstrate final mean average precision (mAP) values ranging from 70 to 90% accuracy, indicating the effectiveness of our approach.