EMSNet: Extremely multi-scale network for salient object detection

In salient object detection, accurately segmenting objects across scales and refining boundaries are crucial challenges. We introduce the Multi-UNet Module (MUM), utilizing multiple U-shaped branches at different scales for extremely multi-scale feature extraction, significantly enhancing detection performance. Our novel boundary-weighted structural loss focuses on boundary areas for more precise segmentation, complemented by a Bi-Weighted strategy for balanced learning across varied object scales. Extensive experiments demonstrate that our method achieves highly competitive results across five mainstream datasets. The source code is released at https://github.com/YinYinOvO/EMSNet.