MuraNet: Multi-task Floor Plan Recognition with Relation Attention

The recognition of information in floor plan data requires the use detection and segmentation models. However, relying on several single-task models can result ineffective utilization relevant when there are multiple tasks present simultaneously. To address this challenge, we introduce MuraNet, an attention-based multi-task model for data. In adopt a unified encoder called MURA as backbone with two separated branches: enhanced decoder branch decoupled head based YOLOX, respectively. architecture MuraNet is designed to leverage fact that walls, doors, windows usually constitute primary structure plan's architecture. By jointly training both tasks, believe effectively extract utilize features tasks. Our experiments CubiCasa5k public dataset show improves convergence speed during compared like U-Net YOLOv3. Moreover, observe improvements average AP IoU respectively.Our ablation demonstrate achieves better feature extraction multi-head branches different further performance. We our proposed disadvantages improve accuracy efficiency recognition.