Lightweight Multi-Scale and Multi-Level Feature Fusion Network For Rgb-D-T Salient Object Detection

Опубликована: Янв. 1, 2025

Most RGB-D-T SOD models are committed to constructing complex feature extraction and multi-modal fusion modules for dealing with scenarios that RGB unable handle. However, they neglect the portability mobility of models. To facilitate practical applications, we introduce LMFNet, a lightweight model. We reduce base channels focus on building multi-level, multi-scale module ensure model can effectively extract most discriminative features across multiple scales, achieving overall low-parameter high-efficiency performance. Concretely, use three parameter sharing encoders from RGB, depth, thermal modalities, which significantly reduces number parameters required. In addition, simplify task, designed an efficient performed in order first fusing depth features, then result image feature. This approach helps risk important information being overshadowed during process. Then, improve model's performance, fuse different levels at varying allowing each layer contain rich detail semantic information.Different incorporate scales. Finally, employ decoder progressively decode multi-level obtained previous stage. this process, fused integrated coarse fine, gradually predicting precise salient object boundaries. Extensive experiments demonstrate our strikes optimal balance between processing speed Specifically, model, only 2.7M parameters, achieves 202 FPS NVIDIA 2080ti 352 × input size while maintaining superior The code is available https://github.com/banjamn/LMFNet.

Язык: Английский

---

Multi-focus image fusion algorithm based on adaptive connection and hybrid convolution attention

Computers & Electrical Engineering, Год журнала: 2025, Номер 123, С. 110236 - 110236

Опубликована: Март 9, 2025

Язык: Английский

---

Dual path transformer with element-wise attention and group cross-aggregation network for medical image segmentation

Computers & Electrical Engineering, Год журнала: 2024, Номер 122, С. 109928 - 109928

Опубликована: Дек. 13, 2024

Язык: Английский

---

Multi-scale target edge uncertainty-aware consistent segmentation network

Computers & Electrical Engineering, Год журнала: 2024, Номер 122, С. 110007 - 110007

Опубликована: Дек. 16, 2024

Язык: Английский

---