Detection of furniture on 2D-floor plans using neural networks

This paper presents the results of applying neural networks for automatic furniture detection on 2D-floor plans. The impact of fine-tuning You Only Look Once (YOLO) models on a custom dataset, collected and annotated as part of the study, was investigated. It was found that the YOLOv9-e model required extensive fine-tuning and showed average recognition accuracy. The results of using the YOLOv11x model, which demonstrated high efficiency and excellent accuracy after minimal fine-tuning, were analyzed. The key advantages and disadvantages of various YOLO model versions for processing 2D plans were identified.