KuShulan Paper Cutting Dataset and Benchmark Experiments Using Object Detection

KuShulan Paper Cutting is a distinctive form of Chinese collage-style paper cutting that integrates cutting, pasting, and layering to produce works rich in visual density and cultural symbolism. Recognized by UNESCO as an important intangible cultural heritage, it features vibrant colors, dense motifs, and abstract forms. However, the visual complexity poses major challenges for conventional object detection models, and no prior work has built a dataset or applied detection techniques tailored to this style. We introduce the KuShulan Paper Cutting Dataset (KSL-PCD), annotated across four semantic categories: person, animal, plant, and daily object. YOLO models are evaluated on the dataset. Larger variants such as YOLOv10x and YOLOv8x achieve the highest detection accuracy, while lightweight models like YOLOv5nu offer faster inference with competitive performance. Nonetheless, all models struggle with small, overlapping, and symbolic elements. These results highlight both the potential of deep learning for cultural heritage applications and the need for architectures adapted to non-photorealistic visual styles. KSL-PCD offers a valuable resource for future work in digital preservation, object recognition, and semantic analysis of folk art.