A Performance Comparison of Balinese Carving Motif Detection and Recognition using YOLOv5 and Mask R-CNN

Balinese carving has various forms of unique motifs that adorn sacred buildings in Bali. As an effort of cultural object preservation, object detection and recognition method are needed to collect Balinese carving objects. However, the object detection and recognition task on Balinese carving is challenging due to the limited dataset available. This study proposed an object detection and recognition method with our currently developing dataset to detect and recognize Balinese carving motifs. We compared and evaluated the performance of YOLOv5 and Mask R-CNN to detect and recognize Balinese carving motifs based on mean average precision, training times, and visualization of detection results. The performance comparison of Mask R-CNN and YOLOv5 aims to produce the best detection and recognition models for Balinese carvings. Based on experiments, YOLOv5 outperformed Mask R-CNN with a mAP@[.5-.95] score of 0.987 and faster training times. However, the advantage of Mask R-CNN adds a mask to the final detection, thus showing better motif segmentation with a mask.