Liquid Fusion of Heterogeneous Representations Towards General Salient Object Detection

2026-06-25

AI Summary

This work addresses the inherent spectral preference discrepancies between convolutional neural networks and state space models, which existing salient object detection methods overlook, thereby failing to effectively exploit their representational complementarity. Inspired by liquid neural networks, the authors propose a state-stimulus fusion paradigm that treats ConvNeXt as an exogenous stimulus and VMamba as an evolving internal state. A dynamic gating mechanism enables content-aware fusion of these heterogeneous features, while a saliency-guided upsampling operator jointly optimizes spectral and spatial information. This approach pioneers the explicit modeling of spectral complementarity in generic salient object detection and supports multimodal extensions. It achieves state-of-the-art performance across five benchmark tasks—RGB, RGB-D, RGB-T, video SOD (VSOD), and video depth-temporal SOD (VDT)—striking a superior balance between accuracy and efficiency.

Abstract

General Salient Object Detection (SOD) aims to identify and segment visually interesting objects from uni-modality or multi-modality scenes, recently advanced by cutting-edge State Space Models (SSMs). However, a critical limitation of current approaches is their neglect of the inherent spectral biases exhibited by different neural network paradigms. By digging to the dataset-level spectral analysis of Convolutional Neural Networks (CNNs) and SSMs, their semantic representations are inherently complementary based on their complementary frequency preferences. Inspired by this, we harmonize heterogeneous representations from SSMs and CNNs to bridge their spectral biases for general salient object detection. To this end, inspired by the dynamic information propagation of Liquid Neural Networks (LNNs), we introduce a liquid fusion to dynamically integrates features from two backbones, including VMamba and ConvNeXt, referred to Liquid Fusion Network (LFNet). Concretely, by treating the continuous VMamba features and ConvNeXt features as evolving states and exogenous stimulus, respectively, LFNet employs a dynamic gating mechanism for content-aware feature aggregation. Crucially, this state-stimulus paradigm enables to scale to multi-modal cues, resulting in flexibility in general SOD. Besides, a Saliency-Guided Upsampling (SGU) operator to propagate the features to the shallow layer, which leverages a spectral-spatial co-design to suppress upsampling artifacts while preserving semantics. Extensive experiments across five diverse tasks (RGB, RGB-D, RGB-T, VSOD, and VDT) demonstrate that LFNet achieves state-of-the-art performance, offering a superior trade-off between detection accuracy and model efficiency. Code has been released at https://github.com/cke520/LFNet.

Problem

• Salient Object Detection
• Spectral Bias
• Heterogeneous Representations
• Multi-modality
• Generalization

Innovation

• Liquid Fusion
• State Space Models
• Spectral Bias
• Dynamic Gating
• Saliency-Guided Upsampling