Chinese Traditional Pattern Hierarchical Annotation Dataset

This dataset comprises 5,700 high-resolution images of Chinese traditional cultural relics with patterns, along with 13,841 bounding box annotations featuring hierarchical semantic relationships. It covers 17 core pattern categories, including Rosette (Baoxianghua), Eight Auspicious Symbols (Bajixiang), ‘Fu’ Character, and ‘Gong’ Pattern. The images are sourced from publicly available digital repositories of authoritative institutions such as the Palace Museum and the National Museum of China, as well as classic academic publications. The annotation was conducted by an interdisciplinary team following the《T/CPRA 200.2-2024》standard. Annotations include pattern category, bounding box coordinates, compositional hierarchy (Primary/Secondary), material type, preservation status, and inter-pattern relationships. The dataset is partitioned into training (70%), validation (15%), and test (15%) sets. It is provided in both YOLO and JSON formats, supporting computer vision tasks such as object detection, image classification, and pattern composition analysis. In benchmark tests, the YOLOv8 model achieved an mAP@0.5 of 82.7% on this dataset.