RSONet: Region-guided Selective Optimization Network for RGB-T Salient Object Detection

Bin Wan; Runmin Cong; Xiaofei Zhou; Hao Fang; Chengtao Lv; Sam Kwong

cs.CVMar 13, 2026

RSONet: Region-guided Selective Optimization Network for RGB-T Salient Object Detection

Authors: Bin Wan, Runmin Cong, Xiaofei Zhou, Hao Fang, Chengtao Lv, Sam Kwong

Organizations: School of Control Science and Engineering, Shandong University, Jinan 250061, China, and also with the Key Laboratory of Machine Intelligence and System Control, Ministry of Education, Jinan 250061, China · School of Automation, Hangzhou Dianzi University, Hangzhou 310018, China · School of information engineering, Huzhou University, Huzhou 313000, China · School of Data Science, Lingnan University, Tuen Mun, Hong Kong

Abstract

This paper focuses on the inconsistency in salient regions between RGB and thermal images. To address this issue, we propose the Region-guided Selective Optimization Network for RGB-T Salient Object Detection, which consists of the region guidance stage and saliency generation stage. In the region guidance stage, three parallel branches with same encoder-decoder structure equipped with the context interaction (CI) module and spatial-aware fusion (SF) module are designed to generate the guidance maps which are leveraged to calculate similarity scores. Then, in the saliency generation stage, the selective optimization (SO) module fuses RGB and thermal features based on the previously obtained similarity values to mitigate the impact of inconsistent distribution of salient targets between the two modalities. After that, to generate high-quality detection result, the dense detail enhancement (DDE) module which adopts the multiple dense connections and visual state space blocks is applied to low-level features for optimizing the detail information. In addition, the mutual interaction semantic (MIS) module is placed in the high-level features to dig the location cues by the mutual fusion strategy. We conduct extensive experiments on the RGB-T dataset, and the results demonstrate that the proposed RSONet achieves competitive performance against 27 state-of-the-art SOD methods.

Figures & tables

Fig. 1: Examples of RGB-T challenge scenarios.

Fig. 3: Illustration of the guidance map.

Fig. 4: Architecture of the context interaction module.

Fig. 5: Architecture of the spatial-aware fusion module.

Fig. 6: Architecture of the selective optimization module.

Fig. 7: Architecture of the dense detail enhancement module.

Fig. 8: Architecture of the mutual interaction semantic module.

	Methods		VT5000				VT1000				VT821
			$\mathcal{M}$ $\downarrow$	$F_{\beta}$ $\uparrow$	$S_{\alpha}$ $\uparrow$	$E_{\xi}$ $\uparrow$	$\mathcal{M}$ $\downarrow$	$F_{\beta}$ $\uparrow$	$S_{\alpha}$ $\uparrow$	$E_{\xi}$ $\uparrow$	$\mathcal{M}$ $\downarrow$	$F_{\beta}$ $\uparrow$	$S_{\alpha}$ $\uparrow$	$E_{\xi}$ $\uparrow$
RGB-T	Methods	MIDD [ 59 ]	TIP 21	.046	.788	.856	.893	.029	.870	.907	.935	.045	.803	.871	.897
	MMNet [ 60 ]	TCSVT 21	.044	.777	.863	.888	.027	.859	.914	.931	.040	.792	.874	.893
	CSRNet [ 61 ]	TCSVT 21	.042	.809	.868	.907	.024	.875	.918	.939	.038	.829	.885	.912
	ECFFNet [ 62 ]	TCSVT 21	.038	.803	.875	.911	.022	.873	.924	.947	.035	.807	.877	.907
	APNet [ 63 ]	TETCI 21	.035	.816	.876	.917	.022	.880	.922	.950	.034	.814	.868	.911

TABLE I: Quantitative comparisons with 27 methods on three RGB-T datasets. The top three results in each column are marked with red , blue , green color, respectively.

\uparrow

and

\downarrow

denote that larger value is better and smaller value is better, respectively.

Fig. 9: Visual comparisons with some methods on RGB-T datasets. The predictions of compared method are listed in each column. (a) RGB images, (b) thermal images, (c) ground truth, (d) Ours, (e) ADF [ 65 ] , (f) MIDD [ 59 ] , (g) MMNet [ 60 ] , (h) CSRNet [ 61 ] (i) ECFFNet [ 62 ] , (j) TAGFNet [ 70 ] , (k) APNet [ 63 ] , (l) CGFNet [ 64 ] , (m) TNet [ 66 ] ,(n) MCFNet [ 34 ] , (o) CCFENet [ 67 ] , (p) CAVER [ 71 ] , (q) SwinNet [ 37 ] , (r) WaveNet [ 72 ] , (s) HRTransNet [ 68 ] , (t) UidefNet [ 69 ] , (u) SPNet [ 73 ]

Method	MIDD	CGFNet	TNet	SwinNet	TAGFNet	LAFB	PATNet	Samba
Speed(FPS)	42.2	52.3	49.6	30.9	39.5	43	10.6	-
Params(M)	52.4	69.9	87	198.8	36.2	453	94.9	54.9
FLOPs(G)	216.7	347.8	39.7	124.7	115.1	139.7	51.1	71.6
Method	WaveNet	DFENet	DSCDNet	SMRNet	ISMNet	SAMSOD	ContriNet	Ours
Speed(FPS)	7.7	11.7	9.6	32.5	6.1	9.4	-	9.4
Params(M)	30.2	149.6	92.3	31.5	114.3	32.7	96.3	88

TABLE II: Comparison of the model parameter, complexity and the average running speed.

Settings	$M$ $\downarrow$	$F_{\beta}$ $\uparrow$	$S_{\alpha}$ $\uparrow$	$E_{\xi}$ $\uparrow$
w/o SO (Addition)	0.0217	0.8883	0.9213	0.9523
w/o SO (Multiplication)	0.0208	0.8948	0.9231	0.9587
w/o SO (Concatenation)	0.0215	0.8896	0.9224	0.9558
w/o SO (Pixel-wise)	0.0203	0.8951	0.9239	0.9605
R $\longrightarrow$ T	0.0215	0.8898	0.9230	0.9561
T $\longrightarrow$ R	0.0216	0.8896	0.9233	0.9554

TABLE III: Ablation study on different component.

Settings	$M$ $\downarrow$	$F_{\beta}$ $\uparrow$	$S_{\alpha}$ $\uparrow$	$E_{\xi}$ $\uparrow$
ResNet-18	0.0408	0.8012	0.8618	0.9052
ResNet-34	0.0358	0.8146	0.8770	0.9125
ResNet-50	0.0413	0.7965	0.8638	0.8994
SAM	0.0438	0.8222	0.8409	0.9074
DINO	0.0372	0.8562	0.8751	0.9258
Ours	0.0197	0.9071	0.9261	0.9632

TABLE IV: Ablation study on different backbone.

Fig. 10: Illustration of the impact of dense detail enhancement module.

Fig. 11: Illustration of the impact of mutual interaction semantic module.

Salient Object Detection

Salient Object Detection finds visually prominent objects in images or video and predicts their pixel-level foreground masks.

See all 37 papers

RA-SOD: Reliability-Aware RGB-T Salient Object Detection under Modality Degradation
Sep 11, 2026Hongbo Gao, Zhengyu Li, Xueru Nie +4Multimodal Robustness Salient Object Detection
State Key Laboratory of Robotics and Systems, Harbin Institute of Technology, No. 92 West Dazhi Street, Nangang District, Harbin, Heilongjiang 150006, China · School of Computer Science, University of Sydney, Camperdown, Sydney, NSW 2006, Australia
DRPFNet: Dual-domain Residual Progressive Fusion Network for RGB-Thermal Object Detection
Aug 4, 2026Zian Wang, Changchun LiMultispectral Object Detection Infrared Object Detection
College of Computer Science and Technology, Jilin University, China
Efficient RGB-T Object Detection via Sparse Cross-Modality Fusion
Jun 29, 2026Chao Tian, Zikun Zhou, Chao Yang +2Multispectral Object Detection Efficient Object Detection
Harbin Institute of Technology, Shenzhen, Guangdong, China

Sep 11, 2026J/K move · Enter open · S save

RA-SOD: Reliability-Aware RGB-T Salient Object Detection under Modality Degradation

Hongbo Gao, Zhengyu Li, Xueru Nie, Dihao Zhu+3

State Key Laboratory of Robotics and Systems, Harbin Institute of Technology, No. 92 West Dazhi Street, Nangang District, Harbin, Heilongjiang 150006, China · School of Computer Science, University of Sydney, Camperdown, Sydney, NSW 2006, Australia

Multimodal Robustness Salient Object Detection Multimodal Fusion

Open paper AskPDF arXiv

RGB-Thermal (RGB-T) salient object detection leverages complementary cues from visible and thermal modalities to improve robustness in challenging environments. However, in real-world scenarios, the reliability of each modality is inherently unstable: RGB images degrade under low illumination, motion blur, and noise, while thermal imagery often suffers from contrast compression and sensor artifacts. Such degradation introduces unreliable perceptual evidence that can mislead cross-modal fusion and significantly deteriorate detection performance. To address this challenge, we propose RA-SOD, a reliability-aware RGB-T salient object detection framework that explicitly models modality reliability and integrates it into feature learning and cross-modal fusion. First, we introduce a reliability-conditioned representation that adaptively compensates degraded modality features while preserving structural cues. Second, an uncertainty-guided dual-stream refinement strategy progressively corrects cross-modal representations while suppressing unreliable evidence. Finally, we propose a pixel-wise modality competition mechanism that dynamically selects modality cues according to spatial reliability for fine-grained fusion. Extensive experiments on four benchmarks (VT821, VT1000, VT5000, and VT-IMAG) demonstrate that RA-SOD achieves state-of-the-art performance and exhibits strong robustness under severe modality degradation. Code and models are available at https://github.com/zaoxienian/RA-SOD.

Abstract

Salient Object Detection

Explore similar work

RA-SOD: Reliability-Aware RGB-T Salient Object Detection under Modality Degradation

DRPFNet: Dual-domain Residual Progressive Fusion Network for RGB-Thermal Object Detection

Efficient RGB-T Object Detection via Sparse Cross-Modality Fusion