Offset-Guided Attention Network for Room-Level Aware Floor Plan Segmentation

Recognition of floor plans has been a challenging and popular task. However, existing approaches are struggling to make accurate room-level unified predictions, seriously limiting their visual quality applicability. In this paper, we propose novel approach recognize the plan layouts with newly proposed Offset-Guided Attention mechanism improve semantic consistency within room. addition, present Feature Fusion module that leverages channel-wise attention encourage room, wall, door further enhancing consistency. Experimental results manifest our is able outperforms works both qualitatively quantitatively. To summarize, paper tackles inconsistent prediction issues in segmentation enhances plausibility, quantitative performance, practical