FANet: focus-aware lightweight light field salient object detection network

The detection and segmentation of salient objects in light field scenes pose significant challenges due to redundant and noisy information. We introduce FANet, a novel light field salient object detection (LF SOD) network. This network leverages focus-aware techniques to enhance both performance and efficiency. FANet employs a focus-aware module (FAM) that merges low-level local information with high-level semantic information. This enables precise capture of contours and boundaries. It also incorporates a lightweight cross-modal analysis module (CMAM), which utilizes a grouped multi-head self-attention mechanism for efficient hierarchical cross-modal analysis. This design significantly improves the model’s ability to distinguish between foreground and background while optimizing computational resources. FANet demonstrates outstanding performance on the HFUT-Lytro dataset, achieving state-of-the-art results. Notably, it operates at 97.4 frames per second on a single NVIDIA RTX 4090 GPU, with only 5.3 million parameters and minimal GPU memory usage. This emphasizes its suitability for real-time applications and showcases an optimal balance between speed and detection accuracy.